Introduction: Obesity is partly driven by unhealthy food choices underpinned by cognitive biases, including approach bias (an automatic tendency to move toward rather than away from appetitive food cues) and delay discounting (a preference for smaller, immediate over larger, delayed rewards). Cognitive training strategies aimed at modifying these biases, namely, approach–avoidance training (AAT) and episodic future thinking (EFT) have been shown to improve food choice. However, previous studies tested these training strategies in single laboratory-based sessions among healthy participants. We conducted a pilot randomised trial to examine the effect of these two trainings, delivered daily for one week via smartphone apps, on approach bias for healthy and unhealthy food, delay discounting, and food choice. Method: Sixty participants with overweight or obesity (39F; age=26.93±6.73 years; BMI=30.34±3.75 kg/m2) were randomly allocated to AAT, EFT, or a waitlist control. Outcomes were measured at pre-training, post-training, and at 6‐week follow‐up. Additional measurements included weight (kgs) and training engagement. Results: Training session completion rates were high for AAT (85.71%) and EFT (86.43%), t(38) = −0.11, p = 0.92. Approach bias for unhealthy food was lower in AAT than EFT at post-training (MDiff = −64.56, p = 0.02, 95% CI [−118.83, −10.28]). Healthy food choice (%) was higher for AAT than controls at post-training (MDiff = 23.45, p = 0.01, 95% CI [7.26, 39.64], d = 1.26), and 6‐week follow‐up (MDiff = 23.92, p = 0.01, 95% CI [5.37, 42.48], d = 1.24), and weight reduced from pre-training to 6‐week follow‐up in AAT (MDiff = −0.74, p = 0.03, 95% CI [−1.40, −0.090], d = 0.47). However, EFT did not affect delay discounting, food choice, or weight (all p's >0.1). Conclusion: AAT is a useful training strategy for improving food choice in obesity and smartphones are a feasible, engaging way to deliver training.