This observational study examined the relationship between mother–child spontaneous joint play and the development of conduct problems in preschoolers, using a short-term longitudinal design. The sample consisted of 60 children showing a range of levels of conduct problems, recruited at their third birthday through community health services in low SES areas. Spontaneous joint play and other mother–child activities were coded from naturalistic, unstructured observations in the home. Amount of time spent in joint play at age 3 predicted individual improvement in conduct problems at age 4, and importantly, this association was independent of initial level of child conduct problems and hyperactivity, social class, maternal depression, and frequency of negative mother–child interactions. The amount of time the child spent unoccupied and not interacting with mother independently predicted worsening conduct scores over time. However, time spent in other activities, including joint conversation and solitary play did not predict change in conduct scores over time. The results suggest that positive and proactive parenting processes such as joint play may make a unique contribution to the very early development of conduct problems, independent of other risk factors. There was no evidence that this association was mediated by child factors such as hyperactivity and poor attention skills.