Pathological gambling is widespread among adolescents (3-8%). Gambling proneness can be evaluated in animals using the Probabilistic Delivery (PD) task. In this operant protocol, rats learn to choose for large over small reward. Subsequently, the probability of large reward-delivery decreases progressively to very low levels. Using a home-cage version of the PD task, we studied (Exp. 1-3) the development of preference for the largest reward in middle (pnd 34-35) and late (pnd 48-49) adolescent rats, using the standard paradigm (Zoratto et al., 2012) and then modifying: (i) probability "p" initially associated with the largest reward; (ii) size difference between rewards; (iii) "removable" or "fixed" partitions (allowing to house animals in couples, separating them only during testing). The standard paradigm (p = 50%, 2 vs 6 pellets; "removable" partitions) does not allow the establishment of preference for the largest reward, at neither adolescent age. Conversely, the modified paradigm (p = 66%; 1 vs 5 pellets; "fixed" partitions) allows the development of such preference, already at pnd 34-35. By using the best combination of these factors, we then investigated (Exp. 4) the characteristics of gambling behaviour in middle adolescent (pnd 36-49) and young adult (pnd 67-80) rats. Gambling proneness appears slightly increased during adolescence when compared to adulthood. Notably, inadequate responses (expressed during post-choice timeout, 30 s) appear markedly reduced, suggesting developing animals to be insensitive to reward-delivery omission. In conclusion, methodological refinement is essential to allow the study of risk-prone behaviour during rat adolescence, thus contributing to a better understanding of psychobiological determinants of gambling.