〔Abstract〕 Objective To establish a laboratory model of heroin addiction in C57BL/6 mice based on associative learning mechanisms. Methods The black box and white box were selected as the memory training environment, and three behavioral training paradigms were studied:(1) Pavlovian conditional position preference(CPP) training paradigm, mice were placed in the white box for memory reinforcement training for 30 min after intraperitoneal injection of 0.1 ml of corresponding concentrations(5.0, 10.0, 20.0 mg/kg) of heroin at 9:00 a.m., and 24 h later 0.1 ml of 0.9% NaCl solution was injected intraperitoneally into the black box for training, and after the training, the mice were tested for their memory preference for the black and white boxes(movement time of different boxes).(2) A naloxone conditional position aversion(CPA) training paradigm was conductedbased on the results of the CPP training paradigm.(3) Behavioral sensitization training paradigm, heroin addiction rating scale was established based on the statistical results of 3 behavioral experiments and the lethality of experimental animal disease mice after drug administration. Three different doses of heroin(5.0, 10.0, 20.0 mg/kg) were selected to induce heroin addiction, and the most appropriate heroin concentration was selected by the results on the rating scale. Results In the CPP training paradigm, CPP was observed in all heroin groups(P<0.05,P<0.001,P<0.05).In the CPA training paradigm, the CPA induction rate was highest in the 10.0 mg/kg heroin group compared to the control group(P<0.01).In the behavioral sensitization training paradigm, all heroin groups caused behavioral sensitization changes(P<0.001);but the 5.0 and 10.0 mg/kg heroin groups did not cause animal mortality. Overall, the 10.0 mg/kg heroin group had the highest dose score on the rating scale. It could be used as a concentration to establish a stable experimental animal model of heroin addiction. Conclusion The study was effective in establishing a heroin addiction model in mice, and it was suitable for modeling drug concentration screening, with high animal survival rate and simple and practical. The combined learning mechanism can effectively shorten the model establishment period.