Analytical models are proposed and investigated to relate errors in spatial registration of multispectral and multitemporal data sets to errors in classification. The underlying application is the classification of crops. The effects of misregistration on the statistics of field-center pixels and boundary pixels are determined. Numerical calculations of the probability of error and probability of correct classification are found for a number of specific cases. Additional effects when more than two classes are present in the sample are also investigated. Elements of a general misregistration model are discussed. Accurate registration of imagery data is a very costly operation. The results of this paper can be used to evaluate the requirements of registration processors.