This paper reports three experiments that aimed to validate the use of state-dependent learning (SDL) as a novel welfare assessment tool to evaluate the hunger state of feed-restricted broiler breeders. In each experiment, birds alternated every 2 days between two food rations: quantitative feed restriction (QFR) and ad libitum access to the same feed (AL). Each food ration was paired with a different, end of day, coloured food reward. It was predicted that the reward associated with hunger (QFR FR) would be preferred to the food reward associated with AL (AL FR) in a subsequent choice test. The SDL preference testing took place after 4 and 8 days of training. Each bird was tested twice (once per food ration fed on the test day). In experiment 1 (pilot, n = 4), birds preferred the QFR-associated reward during both tests (mean (±S.E.M.) preference: QFR FR: 35.0 (±3.5) g; AL FR: 2 (±1.3) g, but differential food reward intake between hunger states during training confounded the results. In experiment two (n = 12) a smaller food reward was used during training to try and equalise intake. The birds preferred the QFR FR in test 1 only (least significant difference (L.S.D.) = 15.08, P < 0.05). The mean (±S.E.M.) consumption in test 1 was: QFR FR: 31.6 (±4.3) g; AL FR: 9.41 (±2.3) g. However, differential reward intake continued to confound the findings. In experiment three (n = 8), the food reward was made more palatable by feeding moist and food reward intake during training was equalised between hunger states. During testing, birds continued to show a significant preference in test 1 only (L.S.D. = 13.73, P < 0.05). It was concluded that SDL-derived preferences observed do exist but are not a robust phenomenon. Therefore, further research is needed to quantify factors influencing SDL development and maintenance before using SDL as a tool to assess hunger in feed-restricted broiler breeders.