OBJECTIVES: Analysis of observational studies often involves the calculation of Confidence Intervals (CI) for incidence rates. This study compared common analytic methods for calculating CI with Poisson Exact CI method using data following a Poisson distribution. METHODS: Nine simulated datasets were created with 100 (small), 500 (medium) and 1000 (large) subjects, one year follow-up for each subject, and number of events following Poisson distribution with true incidence rates of 5, 10, and 20 per 100 subject-year of follow-up. Three analytic methods were used for calculating CIs of incidence rates: (1) Poisson exact CI, (2) Regression model using SAS GENMOD procedure, (3) Non-parametric method described in Xu and LaValley (2012). RESULTS: The CIs from the three methods largely overlapped with the differences being smaller as the sample size increased.  For a true incidence rate of 5 per 100 subject-year follow-up, the Poisson exact CI, regression model, and non-parametric methods yielded CI of (1.089 - 10.242), (1.501 - 10.658) and (0.160 - 7.840), respectively, for a study size of 100 subjects. The CIs were (2.133 - 5.689), (2.268 - 5.714) and (1.974 - 5.226) for study size of 500 subjects, and were (3.711-6.592), (3.790-6.597) and (3.655-6.345) for study size of 1000 subjects, respectively. Similar trends were found for incidence rates of 10 and 20 per 100 subject-year follow-up. CONCLUSIONS: The three methods yield very similar CIs when the study size is large. However, when the study size is small, both the regression and the non-parametric methods produce biased estimates as compared to Poisson Exact CI. The underlying incidence rate does not noticeably affect the property of the CIs.