The improvement of computational prediction method for protein-protein interactions (PPI) through genome comparisons is an important issue in bioinformatics. The phylogenetic profile method [2] has previously been developed for predicting protein functions and discovering specific PPI. This method basically stems from an assumption that interacting protein pairs are likely to have similar patterns of gene inheritance under the evolutionary restriction depending on PPI. Consider (n+1) different organisms whose genomes are completely sequenced. Suppose that one of the organisms has m genes in the genome, presence or absence of the orthologous gene for each of the m genes is examined for the remaining n organisms. If we assign an integer 1 to the presence and in integer 0 to the absence, we can construct an n-dimensional bit vector for a gene, each element of which corresponds to one of the n organisms. The bit vector is called a phylogenetic profile of the gene. When a pair of genes shares the same or similar bit patterns in their phylogenetic profiles, the gene products are predicted to interact each other. One of the recent progresses for phylogenetic profile method is the development of real-valued profiles [1]. The profiles are represented by continuous numerical values such as sequence alignment scores or its p-values instead of binary values. This method enables us to evaluate the similarity between profiles as Pearson’s correlation coefficient. When a correlation coefficient matrix is given, the detailed information about the interactions between the variables is obtained by calculating the partial correlation coefficient matrix. In our previous work, we have shown that partial correlation coefficient is a useful statistical measure for predicting PPI with high accuracy [3]. In this paper, we developed a new method to predict PPI by using the partial correlation coefficients in order to improve the accuracy of the phylogenetic profile method based on the continuous numerical values. The ability of our method to predict PPI was tested by using real-valued phylogenetic profile constructed from the comparison of completely sequenced genomes. The result suggested that our method could improve the accuracy of the prediction of PPI.