Confidence intervals for the attributable risk in various epidemiologic study designs are obtained, via a transformation, from the confidence interval for the natural logarithm of the product of the probability of being exposed to the risk factor, and the risk ratio minus one. When the estimated attributable risk is between .21 and .79, the width of the logarithmic transformation (LT)-based interval is less than that for a maximum likelihood (ML)-based interval. This simple sufficient condition applies to all three well-known epidemiologic study designs. Computer simulation results further demonstrate the superiority of the LT-based interval to the ML-based one when the sufficient condition is satisfied.