Dual-process models integrate deliberative and impulsive mental systems, and predict dietary behaviours better than deliberative processes alone. Computerized tasks such as the Go/No-Go, Stop-Signal and Approach-Avoidance have been used as interventions to directly alter impulsive behavioural antecedents. The current meta-analysis examines the effects of these tasks on dietary behaviours, explores sample- and task-related characteristics as potential moderators of effectiveness, and examines implicit bias change as a proposed mechanism. Nineteen randomized controlled trials testing one implicit bias intervention (38 comparisons) were included in a random-effects meta-analysis, which indicated small cumulative effects on eating-related behavioural outcomes (g = -0.18, CI95 = [-0.32; -0.05], p = .008) and implicit biases (g = -0.30, CI95 = [-0.50; -0.09], p = .004). Task type moderated these effects, with Go/No-Go tasks producing larger effects than Stop-Signal or Approach-Avoidance tasks. Effects of interventions on implicit biases were positively related to effects on eating behaviour (B = 0.46, CI95 = [0.11; 0.82], p = .01). Future research should focus on Go/No-Go tasks for altering dietary behaviour via the impulsive system, and should explore effects of implicit bias interventions over longer periods of time with repeated exposures, especially in real-world as opposed to laboratory settings.