Abstract: Because of the effects of mixed pixels and plants with the same spectral character as winter wheat, the traditional way of sampling which is stratified by planting scale can not guarantee the estimation accuracy when estimating wheat areas with complicated planting structure by remote sensing. To solve this problem, a comprehensive structural index in sampling was defined which considered the two effects. The experimental data were the TM and QuickBird images of the same area acquired at nearly the same time. The area of winter wheat was estimated in different sampling methods and the standard error, degree of accuracy as well as coefficient of variation of the result were calculated to compare among random sampling, stratified sampling stratified by planting scale and stratified sampling stratified by structure. The results shows that no matter which estimate method is taken or how many samples are chosen, taking planting structure into account can always improve the quality of samples and raise the accuracy, especially when the sample size is small (on which occasion standard error is reduced by 2.0×105 m2 and degree of accuracy is increased by 1%). In this way, this research provided theoretical basis for monitoring the area of winter wheat by using remote sensing images in large scale.