OBJECTIVES: Early prosthetic joint infections (PJIs) are managed with debridement, implant retention and antibiotics (DAIR). Our aim was to evaluate risk factors for failure after stopping antibiotic treatment. METHODS: From 1999 to 2013, early PJIs managed with DAIR were prospectively collected and retrospectively reviewed. The main variables potentially associated with outcome were gathered, and the minimum follow-up was 2 years. For the present study, only patients who were in remission after one debridement and without long-term antibiotic suppression were included. The primary endpoint was implant removal or the need to reintroduce antibiotic treatment due to failure. RESULTS: One-hundred-and-forty-three patients met the inclusion criteria. The failure rate after a median duration of oral antibiotic treatment of 69 days (IQR 45–95 days) was 11.8%. In 92 cases, PJI was due to Gram-positive microorganisms, in 21 cases PJI was due to Gram-negative microorganisms and in 30 cases PJI was due to a polymicrobial infection with both Gram-positive and Gram-negative microorganisms. In Gram-positive infections, rifampicin administered in combination with linezolid, co-trimoxazole or clindamycin was associated with a higher failure rate (27.8%, P = 0.026) than that in patients receiving a combination of rifampicin with levofloxacin, ciprofloxacin or amoxicillin (8.3%) or monotherapy with linezolid or co-trimoxazole (0%). Among patients with a Gram-negative infection, the use of fluoroquinolones was associated with a lower failure rate (7.1% versus 37.5%, P = 0.044). CONCLUSIONS: The only factor associated with failure was the oral antibiotic selection, not the duration of treatment. Linezolid, co-trimoxazole and clindamycin, but not levofloxacin, serum concentrations are reduced by rifampicin; a fact that could explain our findings. Further studies monitoring serum concentration could help to improve the efficacy of these antibiotics when administered in combination with rifampicin.