This study assessed internal reliability and test-retest reliability of attention bias scores for food derived from the dot probe task. A visual dot probe task with food and non-food pictures (presented for 3000 ms) was administered to 53 healthy women on two occasions. Attention bias scores for food were calculated based on manual response latencies (reaction time bias) and concurrent assessment of eye-movements (direction bias and dwell-time bias). Subjective hunger and blood glucose levels were measured on both testing occasions. Dietary restraint and eating disorder symptoms were assessed during the second session. Results showed that direction bias had poor internal and test-retest reliability. Dwell time bias had excellent internal and acceptable test-retest reliability. Reaction time bias had acceptable internal and good test-retest reliability. Exploratory correlational analyses found that hunger, blood glucose, dietary restraint and eating disorder symptoms were not consistently significantly correlated with indices of attention bias for food. Overall, these findings contradict previous studies that reported low reliability of attention bias indices derived from the visual dot probe task. The implications are that a longer stimulus presentation time (i.e. ≥ 3000 ms), the use of eye-tracking and the use of appetizing stimuli can yield reliable attention bias scores for food. However, the interpretation of dot-probe scores of attention bias for food based on a dot probe task with 3000 ms presentation time and the score's relationship to theoretically relevant constructs such as hunger, eating restraint and eating disorder symptoms, require further clarification.