Model-based analysis of skin conductance responses (SCR) can furnish less noisy estimates of sympathetic arousal (SA) than operational peak scoring approaches, as shown in previous work. Here, I compare two model-based methods for analysis of evoked (stimulus-locked) SCR, implemented in two software packages, SCRalyze and Ledalab, with respect to their sensitivity in recovering SA. Four datasets are analysed to compare predictive validity, i.e. the sensitivity to distinguish pairs of SA states that are known to be different. SCRalyze was significantly better able than Ledalab to recover this known difference in four out of five tested contrasts and comparable in the remaining one. SCRalyze performed significantly better than conventional analysis in all contrasts. I conclude that the model-based method engendered in SCRalyze is currently the best available approach to provide robust and sensitive estimates of sympathetic arousal.