Background Primary immune thrombocytopenia is a severe bleeding disorder. About 50-85% of patients achieve initial remission from first-line therapies, but optimal second-line treatment remains a challenge. All-trans retinoic acid (ATRA) has an immunomodulatory effect on haemopoiesis, making it a possible treatment option. We aimed to evaluate the efficacy and safety of ATRA plus danazol versus danazol in non-splenectomised patients with corticosteroid-resistant or relapsed primary immune thrombocytopenia. Methods We did a multicentre, randomised, open-label, phase 2 study of adult patients (>= 18 years) with primary immune thrombocytopenia from five different tertiary medical centres in China. Those eligible were nonsplenectomised, resistant to corticosteroid treatment or relapsed, and had a platelet count less than 30 x 10(9) per L. Masked statisticians used simple randomisation to assign patients (1:1) to receive oral ATRA (10 mg twice daily) plus oral danazol (200 mg twice daily) or oral danazol monotherapy (200 mg twice daily) for 16 weeks. Neither clinicians nor patients were masked to group assignments. All patients were assessed every week during the first 8 weeks of treatment, and at 2-week intervals thereafter. The primary endpoint was 12-month sustained response defined as platelet count of 30 x 10(9) per L or more and at least a doubling of baseline platelet count (partial response), or a platelet count of 100 x 10(9) per L or more (complete response) and the absence of bleeding without rescue medication at the 12-month follow-up. All randomly allocated patients, except for those who withdrew consent, were included in the modified intention-to-treat population and efficacy assessment, and all patients who received at least one dose of the study agents were included in the safety analysis. Study enrolment was stopped early because the trial results crossed the interim analysis efficacy boundary for sustained response. This trial is registered with ClinicalTrials.govnumber NCT01667263. Findings From June 1, 2012, to July 1, 2016, we screened 130 patients for eligibility; 34 were excluded and 96 were randomly assigned. 93 patients were included in the modified intention-to-treat analysis: 45 in the ATRA plus danazol group and 48 in the danazol group. At the 12-month follow-up, sustained response was achieved more frequently in patients receiving ATRA plus danazol than in those receiving danazol monotherapy (28 [62%] of 45 vs 12 [25%] of 48; odds ratio 4 . 94, 95% CI 2 . 03-12 . 02, p=0 . 00037). Only two grade 3 adverse events were reported: one (2%) patient receiving ATRA plus danazol with dry skin, and one (2%) patient receiving danazol monotherapy with liver injury. There was no grade 4 or worse adverse event or treatment-related death in either group. Interpretation Patients with primary immune thrombocytopenia given ATRA plus danazol had a rapid and sustained response compared with danazol monotherapy. This finding suggests that ATRA represents a promising candidate for patients with corticosteroid-resistant or relapsed primary immune thrombocytopenia.