Because it is unrealistic to assess fishes in all locations, models are needed to characterize the nursery habitats of flatfish species. Descriptive habitat models of species presence and categorical analysis regression tree (CART) models of species abundance have been developed in previous studies. Based on collections around Kodiak Island, Alaska in 1991 and 1992, these models have been developed for: age-0 flathead sole (Hippoglossoides elassodon); age-0 Pacific halibut (Hippoglossus stenolepis); age-1 yellowfin sole (Pleuronectes asper); and age-0 rock sole (Pleuronectes bilineatus). In 1995, collections were made in bays along the Alaska Peninsula (an area never before sampled for juvenile flatfishes) and were compared with the previous models developed for Kodiak Island in 1991 and 1992. Very simple descriptive models for Kodiak Island (Norcross et al., 1995, Neth. J. Sea Res. 34: 161–175) accurately predicted the abundance along the Alaska Peninsula of flathead sole (78%), Pacific halibut (96%), yellowfin sole (75%) and rock sole (99%) in specific depth ranges and on specific substrate types. More complex CART models of species abundance (Norcross et al., 1997, Fish. Bull. (US) 95: 504–520) were more precise but not as accurate as the descriptive models because some parameters were not always available at the test locations. Flathead sole were found at temperatures ≤ 8.9°C on mud and mixed mud substrates in similar proportions in 1995 (66%) as in 1991–92 (71%). Similarly, Pacific halibut were ≤ 7.9 km inside bays and at depths ≤ 40 m in 93% of the sites of this study, compared with 89% previously. Seventy-five per cent of yellowfin sole were at depths ≤ 28 m on mixed substrates in both study periods. Rock sole were found on sand and muddy sand substrates at temperatures above 8.7°C in 52% of the sites in present and 69% of the sites in previous studies. This field test demonstrated that both descriptive and CART models were very useful at identifying juvenile habitat parameters, but are insufficient for areas not previously sampled. The resource-selection models verified the importance of a subset of parameters used in earlier models and provided a statistical means (78–87% correct) for prediction of fish distribution in similar areas of the eastern North Pacific.