There are two kinds of applications of principal component analysis (PCA) to analyze population substructures of genetic polymorphism data. One application is for an individual covariance matrix, and the other application is for a marker covariance matrix. The former method is already implemented in EIGENSTRAT 1; the latter method, however, is not common because it cannot be applied, if data include missing typing data (allele call). Here, we describe some modification of a Mixture Model 2, so that it can handle data with missing allele calls (we call it a compensated mixture model (CMM) protocol). MM applies PCA to a marker covariance matrix before applying the normal-distribution mixture model.