This in vivo study aimed to compare the performance of different methods of approximal caries detection in primary molars. Fifty children (aged 5–12 years) were selected, and 2 examiners evaluated 621 approximal surfaces of primary molars using: (a) visual inspection, (b) the radiographic method and (c) a pen-type laser fluorescence device (LFpen). As reference standard method, the teeth were separated using orthodontic rubbers during 7 days, and the surfaces were evaluated by 2 examiners for the presence of white spots or cavitations. The area under the receiver-operating characteristics curve (A_z) as well as sensitivity, specificity and accuracy (percentage of correct diagnosis) were calculated and compared with the McNemar test at both thresholds. The interexaminer reproducibility was calculated using the intraclass correlation coefficient (ICC-absolute values) and the kappa test (dichotomizing for both thresholds). The ICC value of the reference standard procedure was 0.94. At white-spot threshold, no methods tested presented good performance (sensitivity: visual 0.20–0.21; radiographic 0.16–0.23; LFpen 0.16; specificity: visual 0.95; radiographic 0.99–1.00; LFpen 0.94–0.96). At cavitation threshold, both LFpen and radiographic methods demonstrated higher sensitivity (0.55–0.65 and 0.65–0.70, respectively) and A_z (0.92 and 0.88–0.89, respectively) than visual inspection sensitivity (0.30) and A_z (0.69–0.76). All methods presented high specificities (around 0.99) and similar ICCs, but the kappa value for LFpen at white-spot threshold was lower (0.44). In conclusion, both LFpen and radiographic methods present similar performance in detecting the presence of cavitations on approximal surfaces of primary molars.