Cross-Priming Amplification (CPA) has been shown to rapidly and effectively detect Mycobacterium tuberculosis (MTB) in sputum samples under isothermal conditions. However, no performance data exist from peripheral-level tuberculosis (TB) clinics in tuberculosis-endemic countries. We conducted a clinical trial at four county-level TB clinics in China to evaluate the effectiveness of the CPA assay. TB suspects were continuously enrolled by a clinician at each clinic. Following informed consent, each patient provided two sputum specimens (spot and morning sputum). Sputum samples were tested by smear microscopy, solid culture and CPA. The National TB reference laboratory (NTRL) collected all culture positive strains and performed 16S–23S rDNA internal transcribed spacer (ITS) sequence analysis for strain identification. Solid culture was used as the gold standard to evaluate the effectiveness of CPA in detecting MTB. A total of 2200 TB-suspected patients were enrolled at the four county-level TB clinics. Compared to solid culture, the sensitivity and specificity of the CPA test for MTB detection within this group was 84.1% (95%CI, 79.5–88.6) and 97.8% (95%CI, 97.1–98.5), respectively, and the sensitivity in smear-negative cases was 59.8% (95%CI, 49.8–69.8). The test failure rate of CPA was 0.8% (32/3918), significantly lower than the 1.7% (106/6138) culture contamination rate.