Our goal was to determine the extent to which recommendations for primary care practice are informed by high-quality research-based evidence, and the extent to which they are based on evidence of improved health outcomes (patient-oriented evidence). As a substrate for study, we used Essential Evidence, an online, evidence-based, medical reference for generalists. Each of the 721 chapters makes overall recommendations for practice that are graded A, B or C using the Strength of Recommendations Taxonomy (SORT). SORT A represents consistent and good quality patient-oriented evidence; SORT B is inconsistent or limited quality patient-oriented evidence and SORT C is expert opinion, usual practice or recommendations relying on surrogate or intermediate outcomes. Pairs of researchers abstracted the evidence ratings for each chapter in tandem, with discrepancies resolved by the lead author. Of 3251 overall recommendations, 18% were graded ‘A’, 34% were ‘B’ and 49% were ‘C’. Clinical categories with the most ‘A’ recommendations were pregnancy and childbirth, cardiovascular, and psychiatric; those with the least were haematological, musculoskeletal and rheumatological, and poisoning and toxicity. ‘A’ level recommendations were most common for therapy and least common for diagnosis. Only 51% of recommendations are based on studies reporting patient-oriented outcomes, such as morbidity, mortality, quality of life or symptom reduction. In conclusion, approximately half of the recommendations for primary care practice are based on patient-oriented evidence, but only 18% are based on patient-oriented evidence from consistent, high-quality studies.