Autoregressive modelling includes a model identification procedure, that is, it is necessary to choose the order of the autoregressive (AR) process that best describes the given finite record (frame) of the signal. Four previously suggested procedures to choose the “best order” of AR processes have been tested: The “first zero crossing” of the autocorrelation function (FZC), the “final prediction error” (FPE), “Akaike's information criterion” (AIC), and the “criterion autoregressive transfer-function” (CAT). It was found that: (i) For more than 98% of the 1280 frames of Doppler signals analyzed the order selected by the various criteria was ten or less. (ii) For the same records of Doppler signals, FPE, AIC and CAT behave in a very similar manner, but the FZC criterion underestimates the order in relation to the others. (iii) For true AR processes, the order selected is frequently different from the true AR order when frames of 64 samples are used. When more samples are used FPE, AIC and CAT tend to select the correct order. (iv) The effect on the spectral estimate of using too high a model order is usually insignificant, while using too low an order can change the estimate more dramatically, that is, overestimating the model order is better than underestimating it.