MEM(Maximum Entropy Method) for power spectral density estimation has been experimentally investigated. It has been known that this method is particularly valuable if the process contains one or more narrow peaks, and is superior to the conventional methods (FFT etc.) because of much more resolution for a given series of data. However the comparison of the methods on the basis of equal resolution is very difficult since that of MEM is not very well defined.
　The results for the deterministic continuous process (sine waves plus white noise) show the problem of the application of Akaike's FPE (Final Prediction Error) on a criterion for choosing the length of the required prediction error filter.
　According to the results for sine waves plus white noise, it follows that the application of Akaike's criterion for the prediction error filter leads to the prediction of the unpredicted amount. But when the signal to noise ratio decreases to a certain extent, the application of that criterion provides the reasonable results.
　Some preliminaly results of the nonstationality of ocean waves are presented as an illustration of the application of MEM.