A multivariate regression, partial least square (PLS) approach was used to optimize a polymerase chain reaction (PCR) method for mixed communities. This approach, in contrast to univariate ones, provided information on the relative influence of the different factors to be optimized, as well as the interactions between factors. Models that predicted the outcome of further optimization were constructed from the initial experiments and verified experimentally. The models constructed were able to predict the outcome of a second set of experiments with high accuracy. PCR-amplification of DNA from environmental samples is often the first step in microbial community fingerprinting. Inhibitors and low cell numbers in the samples can cause problems with yield, for which compensation is normally made by increasing the number of cycles in the PCR-amplification reaction. Increasing the number of cycles, however, can cause other problems such as heteroduplex formation and increased bias. To avoid these problems the effects of different times of denaturing, annealing, and extension on yield were investigated for 2 different samples, one that consisted of a mixture of 9 laboratory strains, and one that represented the microbial community from the surface of the red alga Delisea pulchra. The multivariate approach showed, in addition to the successful optimization of yield, that the different factors affected the PCR depending on sample type. Annealing time had the largest effect on yield for the mixture of laboratory strains, whereas extension time was most important for the D. pulchra community. We suggest that multivariate optimization is a useful tool for PCR optimization and can be used irrespectively of the particular factors that are being investigated.