This paper describes an artificial neural network response surface methodology (ANNRSM) for process modeling and optimization. The process chosen is that of chemical vapor deposition (CVD) of silicon in a barrel reactor. A desired performance requirement of the barrel CVD reactor is that the deposited layers be uniform in thickness. For modeling this d process, experiments are first planned and conducted following the design of experiments (DOE) methodology. The resulting experimental data are mapped with an artificial neural network (ANN). ANNs with different configurations are systematically trained in a "simple to complex" order by a back-propagation training procedure. Another set of designed experimental data is used to test the predictive accuracy of the ANNs and to identify the network with optimum configuration of the networks. The selected model, ANN response surface, in conjunction with a gradient search scheme is used to locate the optimum settings. The results of using this methodology in identifying optimal settings in the presence of noise are also presented. Experiments performed on a mock-up CVD reactor support the optimum settings obtained using the ANNRSM. A comparison between ANNRSM and regression RSM, shows that ANNRSM is able to build an accurate global model and find the optimum using fewer data especially when the data are noisy.