Doctors' communication with patients is commonly hampered by lack of training in this core skill. This study aimed to assess the efficacy of an intensive 3-day training course on communication skills in a randomised controlled trial with a two-by-two factorial design and several outcomes. 160 oncologists from 34 UK cancer centres were randomly allocated to four groups: written feedback followed by course; course alone; written feedback alone; and control. At each of two assessment periods, consultations with six to ten consecutive, consenting patients per doctor were videotaped. 2407 patients participated. Outcome measures included objective and subjective ratings made by researchers, doctors, and patients. The primary outcomes were objective improvements after the intervention in key communication skills. Course content included structured feedback, videotape review of consultations, role-play with simulated patients, interactive group demonstrations, and discussion led by a trained facilitator. In Poisson regression analysis of counts of communication behaviours, course attendance significantly improved key outcomes. The estimated effect sizes corresponded to higher rates of use of focused questions (difference between course attenders [n=80] and non-attenders [n=80] 34%, p=0.003), focused and open questions (27%, p=0.005), expressions of empathy (69%, p=0.003), and appropriate responses to patients' cues (38%, p=0.026), and a 24% lower rate of use of leading questions (p=0.11). There was little evidence for the effectiveness of written feedback. The communication problems of senior doctors working in cancer medicine are not resolved by time and clinical experience. This trial shows that training courses significantly improve key communication skills. More resources should be allocated to address doctors' training needs in this vital area.