BACKGROUND: Conducting a pragmatic randomised trial in primary care is often accompanied by practical problems. Such problems are seldom reported and may constitute useful lessons for researchers planning future trials. AIM: To address the difficulties involved in patient recruitment and to present measures to minimise bias during outcome assessment. METHOD: A recently conducted trial comparing the effects of corticosteroid injections and physiotherapy for painful stiff shoulder was used to illustrate problems related to patient recruitment and outcome assessment. RESULTS: Recruitment of patients was not without difficulties despite careful preparation. Recruitment was discontinued after 20 months, when 109 of the intended 120 patients had been admitted to the trial. The shoulder trial mainly included patient-oriented subjective outcome measures. Subgroup analyses demonstrated that patient preferences might have had some influence on outcome. CONCLUSIONS: General practitioners might be willing or unwilling to participate in research for many reasons. The researcher should take these motivations into account when inviting physicians to take part in research. Strategies to enhance enrolment should be prepared before the start of the trial. When blinding of patients is problematic, patient preferences should be assessed before randomisation and their influence on the outcome studied. Although involving a blinded independent observer enables a more objective assessment of outcome, the success of blinding should be clearly evaluated.