Objects with sharp contours are preferred less than objects with smooth contours, as sharp angles are thought to be an indicator of threat. In two experiments, we probe the link between low level visual features, such as contour curvature, and affective ratings. In Experiment 1, we used artist-traced line drawings of all images from the International Affective Picture System (IAPS) image set. We computationally extracted the contour curvature, length, and orientation statistics of all images, and explored whether these features are predictive of emotional valence scores. Our results replicate previous research, finding a significant negative relationship between high curvature (i.e., angularity) and emotional valence (p = 0.012). Additionally, we find that length is positively related to valence, such that images containing long contours are rated as more positive (p = 0.049). In Experiment 2, we composed new, content-free line drawings of contours with different combinations of length, curvature, and orientation values. Sixty-seven participants were presented with these images on Amazon Mechanical Turk (MTurk) and had to categorize them as positive or negative. A linear mixed effects model revealed that low curvature, long, horizontal contours predicted participants’ positive responses, while short, high curvature contours predicted participants’ negative responses. Taken together, these findings have implications for theories of threat detection such as the Snake Detection Theory, which posits that humans evolved to fear snakes and thus our thalamic nuclei are able to detect snakes rapidly and automatically. It is unlikely, however, that the thalamus has a true representation of “snake”. Our findings suggest a more plausible scenario, whereby visual features associated with threatening stimuli, such as snakes, are quickly detected and passed on to visual cortex for further processing. We have also identified the low-level contour features that are associated with positive valence.