A membrane thickness process control expert system of chemical vapor deposition (CVD) based on neural network is presented. In general, there are many factors would influence the membrane quality. Most of them can be adjusted by changing the recipe, which are the process parameters of the working machines. Finding out a suitable and steady recipe and on-line real-time controlling the recipe is the target that process engineers devote to. Generally speaking, the recipe adjustment is based on the accumulation of experiences or learning from the try and error results. However, the process of thin film deposition is a very complicate and nonlinear system. It is very difficult to find out the relationships between the variation of process parameters and membrane quality. Therefore, a system was developed to simulate the CVD's process using a technique of neural network. An expert system was then set up by extracting out the regular rule between process input and output from the trained neural network, which would provide references to engineers for the need of on-line recipe adjustment.