A statistical post-processing technique is developed and tested to reduce the Navy global model (NOGAPS) track forecast errors for western North Pacific tropical cyclones during 1992-1996. In addition to the basic storm characteristics, the set of 42 predictors includes various track segments in the 00-72 h NOGAPS forecast as well as a 00-36 h backward extrapolation that is compared with the corresponding best-track positions. Although a NOGAPS forecast to at least 36 h is required to calculate the critical backward predictors, a reduced set of forward predictors that did not include the 48- and 72-h NOGAPS positions still produced the same improvement in track forecasts. Separate sets of statistical regressions are developed and tested for three subsets of the synoptic pattern/region combinations defined by Carr and Elsberry. For cyclones in the standard/dominant ridge combination, the improvement relative to NOGAPS is 61% after 12 h, and remains 8% after 72 h. For cyclones in the poleward/ poleward-oriented pattern/region, the improvement over NOGAPS is 55% after 12 h, and 6% after 72 h. For a combination of cyclones in all remaining pattern/ regions, the improvement relative to NOGAPS is 61% after 12 h, and 10% after 72 h. Comparison of these subsets with a single set of regression equations for all synoptic combinations showed no advantage obtained from using separate equation sets, so the single set is recommended. An independent test with all available 1997 NOGAPS forecasts decreased forecast track error by 50, 22, 12, 9, and 6% at 12, 24, 36, 48, and 72 h.