In this paper we developed an approach using time series Normalize Difference Vegetation Index （NDVI） derived from SPOT VGT for crop yield predicting in American during a five-year span （1998—2002）. In order to remove cloud and extract the characteristics of the vegetation dynamics

the Harmonic Analysis of Time Series （HANTS） algorithm was used on the time series of NDVI image.To exploit effectively the time series of NDVI

linking them as much as possible to crop growing conditions

indicators which can be related closely to crop yield were extracted and used for building the predicting models. The weight average method was used to extract crop growth profile with land cover and SPOT Vegetation data. And then indicators were retrieved from the crop growth profiles

including ascend speed

maximum

descend speed

accumulative total before maximum and accumulative total after maximum. At the mean time

the time series of winter wheat yield are processed using a linear upward trend function in 1980 to 2002 to reduce the tendency of the yield. The weather yield is the difference of the actual yield and the trend yield. The weather yield will be predicted with remote sensing indicators. The weather yield and corresponding indicators are regressed. Only those indicators with high correlation coefficient are selected. The wheat yield are the summary of weather yield and the trend yield. The model was used to predict winter wheat yield in America. The difference is about -11.4% to 7.01% by comparing with USDA NASS data. And the relative coefficient between predicting yield and NASS yield is 0.89.