OBJECTIVES: To determine the impact of ignoring population heterogeneity when Markov models are used to estimate cost effectiveness ratios. METHODS: We constructed a simple Markov model with three health states: healthy, sick, and dead. We modeled heterogeneity by assuming that there are two risk groups (high- and low-risk) differentiated by their probability of getting sick. We used the Markov model to estimate the incremental cost effectiveness ratio (ICER) of a medical intervention that reduces the probability of becoming sick among high-risk individuals but cannot be targeted to a single risk group. We used two methods to estimate the ICER for the intervention: 1. pooling the two risk groups together into a single homogeneous population group; and 2. analyzing the two heterogeneous groups separately. RESULTS: We prove algebraically that the pooled model always overestimates the number of quality adjusted life years of survival (QALYs) gained as a result of a medical intervention. The pooled model may overestimate or underestimate the incremental costs depending on the relationships between several parameter values. Thus, the pooled model may overestimate or underestimate the ICER of an intervention. If all costs vary directly in the amount of time spent in each health state, then both methods always yield the same ICER. These results can be extended to Markov models with more than three health states. CONCLUSIONS: Ignoring heterogeneity may lead to erroneous ICER estimates when Markov models are used to represent disease progression. Since the sign of the error depends on several parameters it may be difficult to interpret comparisons of the results of modeling studies. These problems may not be alleviated by selecting conservative parameter estimates. The simplifying assumption of combining heterogeneous population groups should thus be avoided. Policy makers should be aware of these results when interpreting ICERs estimated using Markov models.