For precisely estimating crop yield by remote sensing data, selection of the optimum phase is one of the key links.?In this study, China HJ-1 satellite CCD images were used for the estimation of three major crops (rice, corn, soybean) in 852 farm of Heilongjiang Province. Through wavelet transform filter and NDVI time series data of auto regressive moving average method, the optimum phase of estimation was established based on smoothed NDVI time series curves of 852 farm. Results showed that the fitting error for rice was -0.003356508, and it was -0.001687117 for corn and -0.002530646 for soybean, respectively. The smoothed NDVI time series curves match the local phenology ideally, and the optimum phase for yield estimation of three major crops can be determined by fitting curves.