Marine ranching is an important measure to achieve marine environmental protection and efficient production of fishery resources. Based on the water environment data of Haizhou Bay marine ranching demonstration area in spring (May), summer (August) and autumn (October) of 2014, selecting permanganate index (COD_Mn), dissolved inorganic nitrogen (DIN), dissolved inorganic phosphorus (DIP), biochemical oxygen demand (BOD) as evaluation indexes, neural network models were developed using BP (back propagation) to evaluate the water environmental carrying capacity of the area. The results showed that the average water environmental carrying capacity index of Haizhou Bay marine ranching in 2014 was higher than 0.6, indicating an ideal level of the carrying status. Water environmental carrying capacity varied seasonally as follows:summer > spring > autumn. The water environmental carrying capacity of marine ranching area was obviously better than that of the control area. The high concentrations of COD_Mn and DIN were the main cause for the slight overload of some stations, which might be related to land-based pollution. In addition, shipping also had a certain negative impact on the water environment. BP neural network model has the advantages of convenient construction, objective and reasonable evaluation results, which could be applied to the study of water environmental carrying capacity in marine pastures and other sea areas.