Before treatment, immediately after treatment and at one week follow up, acoustic probes eliciting eyeblink startles were presented to 37 female spider phobics during a behavioral approach test (BAT). To obtain control startles, the subjects also carried out a BAT with a block of wood (neutral foreground) and a BAT with tasty food items (positive foreground). During the pretreatment assessment eyeblink startle magnitudes did not significantly differ among the three BAT conditions. For all conditions, startle responses were larger during the pre- than during the post-treatment and follow-up assessments. This possibly reflects a general fear-induced startle potentiation during the pretreatment session, due to subjects' anticipating exposure to spiders. At one week follow up, the expected linear trend between affective valence (BAT conditions) and startle magnitude emerged, despite the fact that at this time spider startles were significantly smaller than those before treatment. The self-reported startles closely mimicked the pattern of eyeblink startle responses. The present study indicates that the startle response might be a fruitful outcome variable, indexing aspects of phobia not covered by the more commonly used outcome measures. Yet, its prognostic properties remain to be established.