Objectives: To compare a lifelike baby doll intervention for reducing anxiety, agitation, and aggression in older people with dementia in long-term care (LTC), with usual facility care; and explore the perceptions of care staff about doll therapy. Method: Pilot, mixed-methods, parallel, randomised controlled trial, with follow-up semi-structured interviews. Thirty-five residents from five LTC facilities in Queensland, Australia were randomised to the lifelike baby doll intervention (three, 30-minute, individual, non-facilitated sessions per week) or usual care. Outcomes were changes in levels of anxiety, agitation, and aggression after the 3-week intervention, and short-term effects at week 1. Following intention-to-treat principles, repeated measure MANOVA was undertaken. Qualitative interviews involved five staff. Results: The doll intervention did not significantly reduce residents’ anxiety, agitation, or aggression when compared to usual care at weeks 3 (primary outcome) and 1 (secondary outcome). However, there was a significant group-by-time interaction for the outcome of pleasure–the doll group showed a greater increase in displays of pleasure at week 3 compared to baseline than usual care (F(1,31) = 4.400, p = 0.044; Cohen’s d = 0.74). Staff perceived benefits for residents included emotional comfort, a calming effect, and providing a purposeful activity. Perceived limitations were that doll therapy may only be suitable for some individuals, some of the time, and the potential for residents to care for the doll at the expense of their health. Conclusions: Doll therapy can provide some residents with enjoyment and purposeful engagement. Further research should focus on understanding the individual characteristics and circumstances in which residents most benefit.