In the field of external trajectory measurement error separation, aiming at building up the model for extracting trend part from measurement data, the method of error separating based on multi-scale decomposition of wavelet transform is proposed. At first the trend part is extracted with multi-scale decomposition of wavelet transform, and then subtracts the measurement value with it, in order to get the difference as random error. Determining the decomposing level of wavelet transform is the crucial problem in this method because the decomposing level has a direct influence on the result of error separating. The calculation of optimal levels of wavelet transforms is given, and then the optimal levels for the radar-measured data of low degree AR model is workout with it. The method is simple to implement, and high in performance, according to the simulation result, the random error of range measuring data is separated effectively.