The present study explores the threat bias for fearful facial expressions using saccadic latency as the response mode, with a particular focus on the role of low-level facial information, including spatial frequency, physical contrast, and apparent, perceived contrast. In a simple localisation task, participants were presented with spatially-filtered versions of neutral, fearful, angry and happy faces. Faces were either composed of naturally-occurring, expression-related differences in contrast, normalised for RMS contrast, or normalised for their apparent, perceived contrast. Together, findings show that saccadic responses are not biased toward fearful expressions compared to neutral, angry or happy counterparts, regardless of their spatial frequency content. Saccadic response times are, however, significantly influenced by the physical contrast of facial stimuli, and the extent to which these are preserved or normalised at the physical (RMS matched) and psychophysical (perceptually matched) level. We discuss the implications of findings for the threat bias literature, and the extent to which image processing can be expected to influence behavioural responses to socially-relevant facial stimuli.