Property mapping is a fundamental component of ontology matching and yet hardly any technique goes beyond the identification of single property matches. However, real data often requires some degree of composition, trivially exemplified by the mapping of FirstName, LastName to FullName. Genetic programming offers an alter-native, but the solution space is so large that the required computation effort would be prohibitive. This paper proposes a two-phase instance-based technique for complex datatype prop-erty matching. In the first phase, the technique computes the estimate mutual informa-tion matrix of the property values to (1) find simple, 1:1 matches, and (2) compute a list of possible complex matches. In the second phase, it applies genetic programming to a much reduced search space to find complex matches. The paper concludes with expe-rimental results that illustrate how the technique works and indicate that the technique obtains better results that those achieved by separately using the estimate mutual in-formation matrix or genetic programming.