MOORE, Circuit Judge.

Appellants JTEKT Corporation and Koyo Corporation of U.S.A. (JTEKT, collectively), and NTN Corporation, NTN Bearing Corporation of America, American NTN Bearing Manufacturing Corporation, NTN-Driveshaft, Inc., NTN Bower Corporation, and NTN-BCA Corporation (NTN, collectively) appeal the final judgment in JTEKT Corp. v. United States, 717 F.Supp.2d 1322 (Ct. Int'l Trade 2010) sustaining the final results of the United States Department of Commerce's (Commerce) eighteenth administrative review of ball bearings from Japan. See Ball Bearings and Parts Thereof From France, Germany, Italy, Japan, and the United Kingdom, 73 Fed.Reg. 52,823 (Dep't of Commerce Sept. 11, 2008). Because Commerce failed to adequately explain why it continues to use zeroing in Administrative Reviews while discontinuing the practice in investigations, we vacate and remand. In this remand, Commerce need not reconsider its model match methodology because we explicitly affirmed its use of the sum of the deviations approach in SKF USA, Inc. v. United States, 537 F.3d 1373, 1380 (Fed. Cir.2008). Further, we find no error in the method that Commerce used to break ties as a part of its model match method.

BACKGROUND

In order to determine an antidumping margin, Commerce must compare sales in the exporter's home market (foreign like product sales) with sales in the United States. 19 U.S.C. § 1677(16). Ideally, Commerce would match sales of identical merchandise. Because this is not always possible, the statute allows Commerce to consider sales of similar merchandise. Id. § 1677(16)(B), (C).

The merchandise in this case is ball bearings. The dispute centers on the method Commerce used to determine what constitutes similar merchandise. In its first fourteen reviews of ball bearings, Commerce used the family model match methodology. Under this methodology, Commerce considered sales of products in the exporter's home market that had the same physical characteristics as the United States sale. Commerce considered eight characteristics: load direction, bearing design, number of rows of rolling elements, precision rating, inner diameter, outer diameter, width, and load rating. Any bearing that shared these eight characteristics with the merchandise sold in the United States was considered part of the family of merchandise. Commerce then averaged the prices of the family of bearings for its dumping calculations.

In the fifteenth administrative review, Commerce changed to a new method for determining similar merchandise called the sum of the deviations method. This method allows Commerce to compare the United States sale to the sales of a single product in the exporter's home market rather than an average of sales of a family of merchandise. The method uses the same eight characteristics, but weighs them differently. The matching sale must be identical on four of the eight characteristics: load direction, bearing design type, number of rows of rolling elements, and precision rating. For the remaining characteristics, Commerce will consider products that differ from the subject merchandise. For each characteristic, Commerce determines a percentage difference between the product sold in the United States and the product sold in the comparison market. The sum total of these percentage differences must be less than 40%. Additionally, the cost of manufacturing the good sold in the United States must be within 20% of the cost of manufacturing the home market good. This difference in manufacturing cost is called a DIFMER. If all of these criteria are met, the United States sale is a match with the foreign sale—and DIFMER is accounted for by adjusting the price of the foreign sold good.

Under this method, it is possible for there to be more than one matching product, thus, Commerce breaks ties between matching products to determine the single most similar sale. To break these ties, Commerce compares the level of trade and contemporaneity of the sales. If the tie remains, then Commerce selects the sale with the lowest DIFMER.

Appellants argued to the Court of International Trade that Commerce's use of the sum of the deviations approach is not supported by substantial evidence, that Commerce should have included an additional characteristic in its analysis, and that Commerce's tie breaking method is improper. The Court of International Trade agreed with Commerce on all issues. It held that our opinions in Koyo Seiko Co. v. United States, 551 F.3d 1286, 1290 (Fed. Cir.2009) (Koyo III) and SKF USA, Inc. v. United States, 537 F.3d 1373, 1379 (Fed. Cir.2008) (SKF II) expressly decided that Commerce is free to use the sum of the deviations methodology. JTEKT, 717 F.Supp.2d at 1329-30. Further, the Court of International Trade determined that it was reasonable for Commerce to refuse to add a ninth characteristic to its analysis because that characteristic, the presence of lubricant, was adequately accounted for in the DIFMER. Id. at 1332-33. Finally, the Court of International Trade determined that Commerce has considerable discretion in breaking ties between similar merchandise and the use of level of trade and contemporaneity over DIFMER is reasonable. Id. at 1339-40.

DISCUSSION

We review the Court of International Trade's determinations de novo, stepping into its shoes and applying the same standard of review. SKF II, 537 F.3d at 1377. We uphold Commerce's determination unless it is "unsupported by substantial evidence on the record, or otherwise not in accordance with law." NSK Ltd. v. United States, 510 F.3d 1375, 1377 (Fed.Cir.2007).

I. Model Match Methodology

NTN broadly argues that the sum of the deviations methodology is not supported by substantial evidence. It argues that the family methodology, used in the first fourteen reviews, provides a more accurate result because it relies on exact matches between United States and foreign sales. NTN contends that Commerce has failed to show evidence that the sum of the deviations approach is more accurate than the family methodology. It argues that there is no support for Commerce's statement that the sum of the deviations approach is "more accurate in that it selects a single most-similar model and results in more price-to-price comparisons." J.A. 1066. NTN notes that the only reason there are more price-to-price comparisons is because the sum of the deviations approach loosens the standard for a "match" by allowing certain characteristics to differ between the merchandise. In sum, NTN argues that because the sum of the deviations approach allows for differences between the matched products, it is necessarily less accurate than the family matching approach and that we should therefore preclude its application. NTN argues that this argument is distinct from the arguments made in SKF II and Koyo III because, in those cases, no party argued that Commerce failed to support its use of the sum of the deviations approach with substantial evidence.

We agree with the government that SKF II forecloses this argument. In SKF II we stated that "we have specifically affirmed changes to model-match methodologies by Commerce where reasonable." 537 F.3d at 1380. In SKF II we considered the exact issue in this case, whether Commerce erred by switching from a family model match to a sum of the deviations approach. We credited Commerce's reasoning for making the change—that superior technology allowed it to perform the more complicated sum of the deviations approach and that by using the sum of the deviations approach, it was able to compare a single sale rather than an averaged group of sales. Id. at 1380-81. We held that this was sufficient justification for Commerce's changed method.

NTN's newly phrased argument—that Commerce must support its method with substantial evidence—does not overcome our previous determination that the sum of the deviations approach is reasonable. NTN argues that the sum of the deviations approach cannot be more accurate than the family model match method. But this argument misunderstands our standard of review. We can not review Commerce's methods for relative accuracy, only for reasonableness. SKF II, 537 F.3d at 1380. As we stated in SKF II, Commerce has provided ample justification for the use of this method and it is therefore reasonable.

II. Bearing Type

As an alternative basis for reversal, NTN argues that Commerce erred by refusing to further break down one of the characteristics in the model—bearing design type. NTN argues that insert bearing design types should be broken down further because customers view different models of insert bearing distinctly. We have held that in order to overturn Commerce's determination of design types, the appellant must show that "Commerce's choice of design types ... was unreasonable." Koyo III, 551 F.3d at 1292. In fact, we addressed this issue in Koyo III and held that Commerce's determination of a single design type for insert bearings was reasonable. In this case, Commerce found that in light of the similarities in price, costs, and design, it was appropriate to group insert bearings into a single category. Specifically, Commerce noted that appellant "has not shown, however, why bearings with these specific physical differences and commercial distinctions cannot be reasonably compared." J.A. 194. As we stated in Koyo III, Commerce did not act unreasonably and its determination is supported by substantial evidence.

III. Ninth Characteristic

Appellant JTEKT argues that Commerce should have included a ninth characteristic in the method—the presence or absence of lubrication. JTEKT argues that lubrication affects the potential application of the ball bearings. But we have held that the potential application of a product is not decisive of its classification. Koyo Seiko Co. v. United States, 66 F.3d 1204, 1210 (Fed.Cir.1995). Further, we have held that Congress has granted Commerce considerable discretion to determine what constitutes "foreign like product" under the statute. SKF II, 537 F.3d at 1379. Thus, we will not second guess Commerce's determination that it selected the appropriate eight characteristics that result in accurate matches for the subject merchandise.

IV. Commerce's Tie-Breaker Methodology

As discussed above, Commerce's method for determining like merchandise can result in more than one model match. Because Commerce's goal is to compare individual products or models, it must break these ties. To break ties, Commerce first looks to the level of trade and contemporaneity of the sales. If this does not break the tie, then Commerce will select the product with the lowest DIFMER.

NTN argues that this process is flawed and that Commerce must elevate DIFMER above level of trade and contemporaneity in its tie-breaking determinations. NTN posits that DIFMER, as a measure of the cost of manufacturing a good, relates to the good's physical characteristics. NTN then notes that level of trade and contemporaneity are characteristics of a particular sale, rather than the physical characteristics of the good itself. NTN argues that we should require that Commerce elevate DIFMER ahead of any characteristics of a particular sale because the physical characteristics of the product are more important than any commercial considerations in determining similar merchandise.

NTN argues that the relevant statutes require this result. It notes that under 19 U.S.C. § 1677b(a), a "fair comparison shall be made between the export price or constructed export price and normal value." NTN then points to the statute that defines "normal value" as the "price at which the foreign like product is first sold ... to the extent practicable at the same level of trade as the export price or constructed export price," citing § 1677b(a)(1)(B). NTN argues that this shows that considerations like "level of trade" can only come into play once Commerce has already determined the foreign like product. Thus, NTN contends that it is contrary to the statute to use level of trade to determine the foreign like product.

NTN also points to the statute that defines "foreign like product" which states:

Under the statute, if a product falls under category A (i.e., is identical in physical characteristics), then there is no need to resort to category B. Thus, NTN argues that the statutory scheme requires Commerce to place DIFMER above level of trade and contemporaneity of sale because DIFMER relates to physical characteristics.

Appellees respond that the Court of International Trade was correct and that Commerce has significant discretion in breaking ties between equally similar products. Appellees argue that NTN's statutory arguments are misguided because the fact that two products are equally similar means that they both qualify as foreign like product, thus, the statute that defines foreign like product cannot dictate the tie breaker. Further, the government argues that, before applying the tie breaker, Commerce has already considered DIFMER because only products with a DIFMER of 20% or less can be considered like merchandise.

We agree with the government that Commerce is within its discretion to break ties between equally similar products using level of trade and contemporaneity and only using DIFMER if ties still remain. The statute is silent as to any tie breaking methodology. We have already determined that Commerce's sum of the deviations approach is proper. Thus, if two products are equally similar under the sum of the deviations method, they both qualify as foreign like product. There is no statutory mandate to tell Commerce how to break the tie. Thus, Commerce may, in its discretion, determine a reasonable tie breaking methodology. We agree with the Court of International Trade that the level of trade and contemporaneity of sales are both relevant to the determination of the most appropriate merchandise, and thus are reasonable to use as a tie-breaker before resorting to the DIFMER. Further, NTN overstates the nature of DIFMER as a physical characteristic similar to an inner diameter of a bearing. As the Court of International Trade stated, "[t]he DIFMER adjustment, although related to differences in physical characteristics ... is not itself a physical characteristic." JTEKT, 717 F.Supp.2d at 1340. We agree with the Court of International Trade that Commerce did not err in its use of level of trade and contemporaneity to break ties before resorting to the DIFMER.

V. Commerce's Use of Zeroing

We have long held that Commerce's practice of zeroing is a reasonable statutory interpretation entitled to deference. See, e.g., Timken Co. v. United States, 354 F.3d 1334, 1342 (Fed.Cir.2004). Zeroing is the practice whereby the values of positive dumping margins are used in calculating the overall margin, but negative dumping margins are included in the sum of margins as zeroes. Dongbu Steel Co. Ltd. v. United States, 635 F.3d 1363, 1366 (Fed. Cir.2011). Historically, Commerce used zeroing in both the initial investigation to determine whether dumping occurred, and in the subsequent administrative reviews of its dumping determination. But this practice has changed. In response to pressure from the World Trade Organization (WTO), Commerce changed its practice with respect to investigations and no longer zeroes in that phase. Id. at 1367.

We addressed Commerce's conflicting treatment of zeroing in investigations and administrative reviews in Dongbu. Due to the procedural posture in Dongbu, Commerce had provided no justification for using zeroing in administrative reviews and not in investigations. In that case, we vacated and remanded in order for Commerce to explain its reasoning for such disparate treatment. Id. at 1373.

NTN argues that Dongbu requires that we vacate and remand this case in order for Commerce to provide reasoning for its current practice. Appellee Timken argues that NTN failed to properly raise this issue on appeal. Further, appellees argue that Dongbu does not require vacatur because Commerce, unlike in Dongbu, provided reasons for its zeroing practice in this case.

As an initial matter, NTN did not waive this argument. NTN focuses a significant portion of its brief to challenging Commerce's use of zeroing. Although NTN acknowledges that we have consistently upheld the practice, NTN nonetheless raised the issue and specifically pointed out the contradiction of using zeroing in administrative reviews, but not in investigations. NTN Appellant's Br. 34. NTN did not have the benefit of the Dongbu opinion before filing its briefs and thus could not have argued that the case requires us to vacate, but it nonetheless preserved the issue on appeal by arguing that Commerce's continuing practice of zeroing in administrative reviews, but not in investigations, is unreasonable.

We agree with NTN that Dongbu requires us to vacate and remand. Appellees argue that, unlike Dongbu, Commerce explained its reasoning for continuing to zero in administrative reviews, but not in investigations. Appellees are correct that Commerce attempted to address the exact issue in this case:

J.A. 173-74 (citations omitted). While Commerce did point to differences between investigations and administrative reviews, it failed to address the relevant question—why is it a reasonable interpretation of the statute to zero in administrative reviews, but not in investigations? It is not illuminating to the continued practice of zeroing to know that one phase uses average-to-average comparisons while the other uses average-to-transaction comparisons. In order to satisfy the requirement set out in Dongbu, Commerce must explain why these (or other) differences between the two phases make it reasonable to continue zeroing in one phase, but not the other. Thus, we vacate and remand in order for Commerce to provide its reasoning.

VACATED and REMANDED.