Estimations of the necessary sample size of retrospective data were done using EpiInfo 6.03 STATCALC (CDC/WHO 1996). It was assumed that there would be 99% accuracy in obtaining and recording answers, 95% confidence and 80% power. Using previously estimated rates of antibiotic prescription (25%), and wanting to detect an inappropriate antibiotic prescription rate of at least as low as 35%, a random sample of 348 retrospective records from each facility was necessary to ensure statistical significance (Hogerzeil et al 1993). To control for seasonal variation, a six-month period including both dry and rainy seasons was selected. Sixty records from each month were then collected for a total of 360 retrospective records per facility. Total retrospective sample size was 5040 records. In addition, 100 or 30 prospective records (busier facilities and less busy facilities, respectively) were collected for each facility, for a total prospective sample size of 856. While insufficient to compare between facilities, the total prospective sample size was sufficient to compare antibiotic prescription rates in 1996 and 1997. The sample size of 30 patient interviews and 30 prescriber/dispenser encounters at each facility was based upon WHO recommendations (WHO/DAP 1993).