For diagnostic systems, features are extracted from signals by the usage of autoregressive (AR) and autoregressive moving average (ARMA) methods and power level-frequency distributions of the signals are demonstrated by statistical features. In the present study, spectral analysis of ophthalmic arterial Doppler signals obtained from different subjects was performed using the AR and ARMA methods and Doppler power spectral density values which contain a significant amount of information about the signal were considered as feature vectors representing the signal. In order to reduce the dimensionality of the extracted feature vectors, statistical processes were performed over the Doppler power spectral density values and input feature vectors of multilayer perceptron neural networks used in classification were selected. Performances of the AR and ARMA methods in the analysis of Doppler signals were determined by examining performances of multilayer perceptron neural networks trained by different algorithms. Total classification accuracies of the constructed networks were demonstrated that multilayer perceptron neural network trained by Levenberg-Marquardt algorithm which used ARMA Doppler power spectral density values as inputs could be used in classification of ophthalmic arterial Doppler signals.