A normal random effects model for log-transformed respirator penetration P values was formulated, where P is the ratio of the contaminant concentration inside the respirator CI to the ambient contaminant concentration outside the respirator Co. The model accounts for within- and between-wearer variability in P. Thirty three published and unpublished studies were located which contained measurements for P from multiple respirator wearers, but only seven studies of negative-pressure air-purifying halfmask respirators (HMR's) and two studies of helmet- and-visor type powered air-purifying respirators (PAPR's) provided sufficient detail. In each study, the model for In P was fit by the method of maximum likelihood, which yielded estimates of the overall mean In P value ul, the within-wearer variance ~2w , and the between-wearer variance ~2B. In the seven HMR studies, the estimated between-wearer variance component contributed, respectively, 0%, 0%, 6.5%, 11 %, 23%, 46% and 51 % of the total variance. The findings indicate that for HMR use, the within-wearer variance tends to dominate the between-wearer variance. In the two PAPR studies, the between-wearer variance component contributed, respectively, 0% and 43% of the total variance. Criteria were specified for the assigned penetration factor APF(P) (the inverse of the assigned protection factor APF), namely, no more than 5% of wearers experience more than 5% of P values above the APF(P) value.