Not all mistakes are alike. We study how “blatant” mistakes -- mistakes which even people with limited expertise confidently recognize as such -- influence algorithm aversion and reliance on decision support systems. Across four experiments, we demonstrate that people are less likely to rely on decision support from algorithms after observing them make blatant mistakes. In Experiment 1, participants avoid an algorithm that makes blatant mistakes, and boosting their expertise attenuates this effect. We then confirm that blatant mistakes matter more than mistakes that are equally wrong but harder to spot, both for adopting a predictive model in a quantitative estimation task (Experiment 2) and a large language model in a reasoning task (Experiment 4). Moreover, participants in Experiments 2 and 4 avoid another person after seeing them make a blatant mistake much like they avoid an algorithm that commits the same mistake. Finally, Experiment 3 suggests that blatant mistakes also contribute to the algorithm aversion in a direct replication of an influential study. Our findings point to blatant mistakes as a key driver of algorithm aversion, underscore the need to account for the nature of mistakes when evaluating decision support systems, and emphasize parallels in the adoption of human and algorithmic decision support.