Many geological observation sets contain discrete-state data, which can be encoded as binary patterns. When there are conditional relationships between the variables, latent class analysis may be applied to subdivide the total sample into latent facies associations, which have local independence in the probability sense. Probabilities of latent facies assignments can be mapped areally as continuous surfaces of implied geological facies. Latent class analysis is rooted in simple probability theory and can be a useful technique in geological applications where observations are descriptive or weakly numerical. The method is illustrated by a latent facies mapping of the Morrison Formation (Upper Jurassic) in the subsurface of west Kansas.