Squid is one of the important economic species in the northwestern Pacific. Using Catch per Unit Effort and V% as the target of central fishing ground and adopting BP artificial neural network, we forecast fishing ground in the northwest Pacific Ocean. The study was based on the data of squid fishing and relevant marine environment factors, including longitude, latitude, SST and SSHA from July to November from 1995 to 2001.The input factor is marine environment factor, the output factors are CPUE and V% and 4-3-1 and 4-2-1 model total 4 kinds models were used to compare which is the best suitable model for fishery forecast. The minimum fitting residual of model is the best one. Result shows that 4-3-1 is the best suitable model for each month, but the best suitable model for July and August is 4-3-1 with output V% and best suitable model for September, October and November is 4-3-1 with output CPUE, the minimum overall average error is 4-3-1 model output V%. Research suggests that there are differences as a center of fishery forecast targets by CPUE and V%, and the 4-3-1 model output V% can be used as forecasting model of squid fishing ground.