Empirical models have been used to estimate primary production based on phytoplankton biomass and light intensity. In this paper, an alterative approach for estimating primary production using neural networks is proposed. The inputs to the neural network are chlorophyll, surface irradiance, sea surface temperature, and day length. The output of the network is the estimated primary production. The back-propagation learning algorithm is used to train the neural network. A single step learning with random presentation sequence is selected as the learning strategy. The data set used for this experiment is extracted form the Ocean Primary Productivity Working Group database. The results show a significant decrease in the mean squared error of the log transformed primary production compared to the estimation obtained using linear model and the vertically generalized production model. The neural network- based models can deal with non-linear relationships more accurately, can effectively include variables that tend to co-vary non-linearly with the output variable, are flexible towards the choice of inputs, and are tolerant to noise. Hence to improve the estimation of primary production, additional parameters can be easily incorporated in the neural network model, even though no a prior knowledge about het effect of these parameters is available. These important features of neural networks make them an ideal candidate for constructing primary production models for both case 1 and case 2 waters.