The high microbiologic diversity encountered in prosthetic joint infection (PJI) makes the choice of empirical antimicrobial therapies challenging, especially in cases of implant retention or one-stage exchange. Despite the risk of dysbiosis and toxicity, the combination of vancomycin with a broad-spectrum β-lactam is currently recommended in all cases, even if Gram-negative bacilli (GNB) might be less represented in late PJI. In this context, this study aimed to describe the microbiologic epidemiology of PJI according to the chronology of infection. This prospective cohort study (2011–2016) evaluated the microbiologic aetiology of 567 PJI according to time of occurrence from prosthesis implantation—early (<3 months), delayed (3–12 months) and late (>12 months)—as well as mechanism of acquisition. Initial microbiologic documentation (n = 511; 90.1%) disclosed 164 (28.9%) Staphylococcus aureus (including 26 (16.1%) methicillin-resistant S. aureus), 162 (28.6%) coagulase-negative staphylococci (including 81 (59.1%) methicillin-resistant coagulase-negative staphylococci), 80 (14.1%) Enterobacteriaceae, 74 (13.1%) streptococci and 60 (10.6%) Cutibacterium acnes. Considering nonhaematogenous late PJI (n = 182), Enterobacteriaceae (n = 7; 3.8%) were less represented than in the first year after implantation (n = 56; 17.2%; p <0.001), without difference regarding nonfermenting GNB (4.6% and 2.7%, respectively). The prevalence of anaerobes (n = 40; 21.9%; including 32 (80.0%) C. acnes) was higher in late PJI (p <0.001). Consequently, a broad-spectrum β-lactam might be useful in 12 patients (6.6%) with late PJI only compared to 66 patients (20.3%) with early/delayed PJI (p <0.001). Considering the minority amount of GNB in late postoperative PJI, the empirical use of a broad-spectrum β-lactam should be reconsidered, especially when a two-stage exchange is planned.