Approach biases to foods may explain why food consumption often diverges from deliberate dietary intentions. Yet, the assessment of behavioural biases with the approach-avoidance tasks (AAT) is often unreliable and validity is partially unclear. The present study continues a series of studies that develop a task based on naturalistic approach and avoidance movements on a touchscreen (hand-AAT). In the hand-AAT, participants are instructed to respond based on the food/non-food distinction, thereby ensuring attention to the stimuli. Yet, this implies the use of instruction switches (i.e., 'approach food - avoid objects' to 'avoid food - approach objects'), which introduce order effects. The present study increased the number of instruction switches to potentially minimize order effects, and re-examined reliability. We additionally included the implicit association task (IAT) and several self-reported eating behaviours to investigate the task's validity. Results replicated the presence of reliable approach biases to foods irrespective of instruction order. Evidence for validity, however, was mixed: biases correlated positively with external eating, increase in food craving and aggregated image valence ratings but not with desire to eat ratings of the individual images considered within participants or the IAT. We conclude that the hand-AAT can reliably assess approach biases to foods that are relevant to self-reported eating patterns.