Aim To determine predictors of changes in amount of cocaine use among regular users outside treatment services. Design Longitudinal study—we estimated the proportion of subjects who increased or decreased cocaine use and assessed possible predictors related to these changes among a street‐recruited cohort of young regular cocaine users (RCU). Setting Three Spanish cities: Barcelona, Madrid and Seville Participants A total of 720 RCU aged 18–30 years not regularly using heroin were recruited in the community during 2004–06 (Itinere Project). Follow‐up interviews (n = 501) were carried out at 12–24 months. Measurements The average amount of cocaine used weekly was calculated taking into account the number of days of use and the usual quantity (g/day). A multinomial logistic regression approach was used to investigate the association between changes in amount of cocaine use (i.e. difference exceeded 33.3% of baseline level) after 12–24 months, and baseline socio‐demographic characteristics, nightlife, patterns of cocaine use and use of alcohol and other psychoactive drugs. Findings Cocaine use baseline average level was 2.14 g/week [95% confidence interval (CI) = 2.02–2.42]. It decreased in 71.5% of subjects and increased in 14.1%. In multinomial analysis, negative associations were found between decreasing cocaine use and high levels of alcohol consumption and using an increasing number of psychoactive drugs. Moreover, low education level, having used cocaine frequently in houses and reporting cocaine binges were associated with increasing cocaine use. Conclusions A street‐recruited cohort of cocaine users in Spain showed a significant reduction in cocaine use over a period of 12–24 months. High consumption of alcohol and increasing use of other psychoactive drugs decreased the probability of reducing cocaine use.