A simple and fast strategy is proposed to evaluate the frequency spectrum of a string of data whose total number may not be of 2(superscript m) and the time intervals between successive data may be non-uniform. The cubic moving least squares method is first employed to separate both the non-sinusoidal and random parts. Next, the Huynh M3A cubic monotonic interpolation is modified and is employed to convert the remaining data into a data string with uniform time intervals and the total data number is exactly 2(superscript m). Finally, a simple FFT algorithm is employed to provide the spectrum with negligible low frequency error.