The aim of this work was to investigate the feasibility of using feed-forward neural networks for system identification of a process with highly non-linear characteristics. A biochemical process was chosen where the microorganism Saccharomyces cerevisiae, a yeast, grows in a chemostat on glucose substrate and produces ethanol as a product of primary energy metabolism. This process is of immense interest to industries worldwide. Three state variables considered for the process are the microbial concentration, substrate concentration and production concentration. The Levenberg-Marquardt method was used to train the neural networks by minimising the sum of squares of the residuals. The inputs to the networks were the three state variables at a time, the process input variables (control variables and disturbances) from that time to the time for which the state variables are to be predicted. This duration was 0.5 hour for the first part of the study, and 0.1 hour for the later part of this work. The characteristic time for the process was about 2.9 hours under normal circumstances. The output of each node was calculated by the logistic (sigmoid) or symmetric logarithmoid activation functions on the weighted sum of inputs to that node. In most cases, the symmetric logarithmoid resulted in lower error square sum values than the sigmoid. The predictions were fairly good in the first part of the work, and were quite accurate in the second part. This work demonstrated that system identification can be performed using feed-forward neural networks. The duration for which the prediction is being performed is limited by the number of nodes in the hidden layer (or weigths), and should be much smaller than the characteristic time for the process.