Pavlovian fear conditioning is widely used as a laboratory model of associative learning in human and nonhuman species. In this model, an organism is trained to predict an aversive unconditioned stimulus from initially neutral events (conditioned stimuli, CS). In humans, fear memory is typically measured via conditioned autonomic responses or fear‐potentiated startle. For the latter, various analysis approaches have been developed, but a systematic comparison of competing methodologies is lacking. Here, we investigate the suitability of a model‐based approach to startle eyeblink analysis for assessment of fear memory, and compare this to extant analysis strategies. First, we build a psychophysiological model (PsPM) on a generic startle response. Then, we optimize and validate this PsPM on three independent fear‐conditioning data sets. We demonstrate that our model can robustly distinguish aversive (CS+) from nonaversive stimuli (CS‐, i.e., has high predictive validity). Importantly, our model‐based approach captures fear‐potentiated startle during fear retention as well as fear acquisition. Our results establish a PsPM‐based approach to assessment of fear‐potentiated startle, and qualify previous peak‐scoring methods. Our proposed model represents a generic startle response and can potentially be used beyond fear conditioning, for example, to quantify affective startle modulation or prepulse inhibition of the acoustic startle response.