On-site angler surveys commonly yield data on fish catch, fishing effort, and their variances for a sample of anglers. The ratio of catch to effort, or catch rate, often is multiplied by an independent estimate of total effort to calculate total catch (with its confidence interval) throughout the fishery. The most frequently used measures of catch rate are the ratio-of-means estimator (mean catch per angler divided by mean effort per angler) and the mean-of-ratios estimator (mean angler's catch rate). Bias and misleading confidence intervals are associated with use of ratio estimators, and the best catch rate measure for estimating total catch has been uncertain. We used statistical theory and simulation modeling to demonstrate that the most appropriate estimator (least bias, truest confidence interval) depends on the method of sampling. The ratio-of-means catch rate is better when anglers are sampled with equal probability at the completion of their trips (as in access point surveys). The mean-of-ratios estimator should be used when anglers are sampled, while still fishing, with probabilities proportional to trip length (as in roving surveys). When either estimator is used, confidence intervals around total catch are strongly influenced by skewness in the distribution of catch rates among anglers. At least 100 anglers or angler parties must be interviewed to attain targeted two-tailed ∝-levels; even then, the two statistical rejection areas (ideally, ∝/2) differ between tails of the distribution. Our results clarify the appropriate use of catch rate estimators with survey data, and they show the desirability of larger sample sizes than are customarily used in angler surveys.