Profile hidden Markov models (HMMs) were used to predict the configuration of secondary alcohols and α-methyl branches of modular polyketides. Based on the configurations of two chiral centers in these polyketides, 78 ketoreductases were classified into four different types of polyketide producers. The identification of positions that discriminate between these protein families was followed by fitting six profile HMMs to the data set and the corresponding subsets, to model the conserved regions of the protein types. Ultimately, the profile HMMs described herein predict protein subtypes based on the complete information-rich region; consequently, slight changes in a multiple sequence alignment do not significantly alter the outcome of this classification method. Additionally, Viterbi scores can be used to assess the reliability of the classification.