OBJECTIVE: To model the relationship between resistance to trimethoprim sulphamethoxazole (TMP-SXT) and outcome of pyelonephritis from an RCT comparing TMP-SXT with ciprofloxacin (Talan, JAMA, 283:1583-1590,2000). METHODS: In the trial cost analysis was limited to a comparison between the two treatment groups, which did not permit analysis of effect modifiers, such as TMP-SXT resistance. A probabilistic model was developed and distributions were assigned to expected costs and probabilities of cure, failure (persistent or recurrent infection) and superinfection (reinfection with a new organism) from the trial data. These distributions, which characterise the second order uncertainty surrounding these outcomes following the trial, were propagated through the model using Monte Carlo simulation in order to generate cost effectiveness acceptability curves. RESULTS: There were 255 patients in the per protocol analysis. The mean cost for patients with treatment failure ($476, n=34) was significantly higher than the cost for patients who were either cured ($307, n=174) or had superinfection ($325, n=11). However, treatment with ciprofloxacin reduced both the probability and cost of treatment failure (the mean difference in treatment cost between success and failure was only $4 in the ciprofloxacin arm compared with $267 in the TMP-SXT arm). The probability that ciprofloxacin was dominant (greater efficacy at lower cost) was >95% at 18% TMP-SXT resistance and remained >95% down to a resistance rate of 10%. These results were sensitive to a small number of hospitalisations (1 ciprofloxacin, 5 TMP-SXT). Taking outpatient costs alone, the threshold value of treatment failure at which ciprofloxacin was 90% likely to be cost-effective ranged from $700 for 18% resistance to $350 for 30% resistance. CONCLUSIONS: The stochastic model produces results that are importantly different from a simple analysis based on point estimates of probability and cost, particularly if the latter assumes that the cost of treatment failure is the same for both treatments.