In order to eliminate the subdivision error of moiré fringe signals in the circular grating measurement process, a parameter identification and deviation compensation method for grating sampled signal is proposed. The method is based on the parameter identification theory of the genetic algorithm and cannot be affected by initial signal model parameters with characteristics of optimization and applicability. The regenerated signal model can fit the original signal well. The periodic signals between two grating pitches are sampled with grating rotating at a constant speed in sampling experiments. From the sampled discrete data, the frequency spectra are obtained and analyzed for establishing the mathematical model of grating signals. The parameters which cause the subdivision error are identified by using the genetic algorithm to compensate the subdivision error. The experimental results show that the parameters of signal model constructed by genetic algorithm are identified accurately. The sine deviation of moiré fringe signals is compensated by comparing the Lissajous figures with and without compensation. The subdivision error detected in one grating pitch decreases from 10.65″ to 3.31″. This method can actually be used in grating encoder system and other displacement measuring systems, which ensures the measurement accuracy and reliability.