Background: South Africa has a large burden of rifampicin-resistant tuberculosis (RR-TB), with 18,734 patients diagnosed in 2014. The number of diagnosed patients has increased substantially with the introduction of the Xpert MTB/RIF test, used for tuberculosis (TB) diagnosis for all patients with presumptive TB. Routine aggregate data suggest a large treatment gap (pre-treatment loss to follow-up) between the numbers of patients with laboratory-confirmed RR-TB and those reported to have started second-line treatment. We aimed to assess the impact of Xpert MTB/RIF implementation on the delay to treatment initiation and loss to follow-up before second-line treatment for RR-TB across South Africa. Methods and findings: A nationwide retrospective cohort study was conducted to assess second-line treatment initiation and treatment delay among laboratory-diagnosed RR-TB patients. Cohorts, including approximately 300 sequentially diagnosed RR-TB patients per South African province, were drawn from the years 2011 and 2013, i.e., before and after Xpert implementation. Patients with prior laboratory RR-TB diagnoses within 6 mo and currently treated patients were excluded. Treatment initiation was determined through data linkage with national and local treatment registers, medical record review, interviews with health care staff, and direct contact with patients or household members. Additional laboratory data were used to track cases. National estimates of the percentage of patients who initiated treatment and time to treatment were weighted to account for the sampling design. There were 2,508 and 2,528 eligible patients in the 2011 and 2013 cohorts, respectively; 92% were newly diagnosed with RR-TB (no prior RR-TB diagnoses). Nationally, among the 2,340 and 2,311 new RR-TB patients in the 2011 and 2013 cohorts, 55% (95% CI 53%–57%) and 63% (95% CI 61%–65%), respectively, started treatment within 6 mo of laboratory receipt of their diagnostic specimen (p < 0.001). However, in 2013, there was no difference in the percentage of patients who initiated treatment at 6 mo between the 1,368 new RR-TB patients diagnosed by Xpert (62%, 95% CI 59%–65%) and the 943 diagnosed by other methods (64%, 95% CI 61%–67%) (p = 0.39). The median time to treatment decreased from 44 d (interquartile range [IQR] 20–69) in 2011 to 22 d (IQR 2–43) in 2013 (p < 0.001). In 2013, across the nine provinces, there were substantial variations in both treatment initiation (range 51%–73% by 6 mo) and median time to treatment (range 15–36 d, n = 1,450), and only 53% of the 1,448 new RR-TB patients who received treatment were recorded in the national RR-TB register. This retrospective study is limited by the lack of information to assess reasons for non-initiation of treatment, particularly pre-treatment mortality data. Other limitations include the use of names and dates of birth to locate patient-level data, potentially resulting in missed treatment initiation among some patients. Conclusions: In 2013, there was a large treatment gap for RR-TB in South Africa that varied significantly across provinces. Xpert implementation, while reducing treatment delay, had not contributed substantially to reducing the treatment gap in 2013. However, given improved case detection with Xpert, a larger proportion of RR-TB patients overall have received treatment, with reduced delays. Nonetheless, strategies to further improve linkage to treatment for all diagnosed RR-TB patients are urgently required.