Scalar implicatures (SIs) such as the inference from 'some' to 'some but not all' are known to vary in processing cost. According to the Polarity Hypothesis (van Tiel et al., 2019; van Tiel & Pankratz, 2021), this cost depends on the polarity of the scalar term: SIs from positively polar expressions (e.g., 'some' implicating 'some but not all') are expected to be slower to compute than literal interpretations, while the reverse should hold for SIs from negatively polar expressions, sometimes referred to as indirect SIs (e.g., 'not all' implicating 'not all but some'). We tested this prediction in a sentence–picture verification experiment with 538 English-speaking participants. Each participant judged whether sentences containing weak or strong scalar terms matched a corresponding picture, allowing us to measure response times as an index of processing effort. Our results showed no clear evidence for the Polarity Hypothesis effect. Specifically, we found that the effect strongly varies across the different scales included in our experiment. We additionally conducted a meta-analysis pooling our estimate with all existing evidence so far: the pooled effect is small and in the predicted direction, but the study-specific estimates behind it vary widely in magnitude. Surprisingly, our results do also not clearly validate the traditional assumption of a B&N effect (Bott & Noveck, 2004), i.e., a general polarity-independent processing slowdown under SIs: a subsequent analysis including trial order revealed that the B&N effect was present in early trials of our study, whereas later trials do not show the effect. Overall, we conclude that polarity may influence the speed of scalar implicature processing, but that this effect is not robust across all types of scales.