Dissolved oxygen （DO） is an important water quality parameter in fishery water. Dissolved oxygen condition has great influence on water quality and growth of cultured organisms. With rapid development of pond cultivation, the dissolved oxygen concentration in pond is gradually attached importance to as factor of water environment. At present we mainly adopt timing and fixed point measurement for dissolved oxygen in pond, so the accurate predication of DO in fishpond has been the key to aquatic breeding. The factor which influences dissolved oxygen in pond is complicated. For a certain pond, the dissolved oxygen is relative to different seasons, measuring time, the position, the depth of tneasturing point, wind speed, the depth and surface area of the pond. The prediction for dissolved oxygen in pond is a problem of multi-variable, non-liuearity and 10ng-time lag. Due to complexity and nonlinearity of influence factor of dissolved oxygen, it is difficult to use precise mathematics model to describe dissolved oxygen quantitatively. The artificial neural network is a nonlinear, optimization tool. By its good characteristics of high nonlinear mapping, self-organization, the ability of high parallel processing, the artificial neural network connects various affected factors. After synthetically analyzing and considering the measurability of all variables, we selected water temperature, nitrite, ammonia value （NO2-N）, and total nitrogen value in the pond as input .variable of the neural network, and dissolved oxygen in pond as output variables of the neural network. This paper applied fuzzy neural networks to predict dissolved oxygen in pond. Fuzzy neural network not only possesses the advantages of the fury system and artificial neural network, but also offsets disadvantages caused by their individual modeling. It collects learning, associating, self-adaptive and fuzzy information processing as a whole. On the basis of this, the project selected fuzzy neural network technology as modeling method of predictiori for dissolved oxygen in the pond. Fuzzy neural networks have nice approximation ability. However, the training of NNs by conventional back-propagation method, i.e. the BPNNs, has intrinsic vulnerable weakness in slow convergence and local minima. Thus it becomes one of the research directions in fuzzy neural network to adopt global searching algorithms to optimize the parameters of fuzzy neural network. A great deal of bibliography adopt genetic algorithm to optimize fuzzy neural network. In this research we adopted the easier global optimization algorithm （particle swarm optimization algorithm） to optimize fuzzy neural network. Particle swarm optimization （PSO） is an evolutionary conaputation technique developed by Kennedy and Eberhart in 1995 and successfully used for nonlinear function optimization and neural network training. It is easy to be achieved and need not adjust lots of pararneters and has characteristics of rapid convergence. In this work, PSO algorithm was applied to training of fuzzy neural network and then compared with BP algorithm, showing faster convergence rate. The experimental results show that the proposed method is effective and more accurate than BP-NNs and the real-world application is potential. The method proposed lays foundation for developing intelligent measuring instrument and applying industrialized mariculture.