This in vitro study of occlusal surfaces on primary molars aimed to: (1) evaluate the reproducibility of the Nyvad and ICDAS-II visual systems in detecting caries; (2) to test the accuracy of the systems in estimating lesion depth, and (3) to examine the association between the Nyvad system and the Lesion Activity Assessment system, an adjunct to ICDAS-II. Two samples of extracted primary molars (sample 1, n = 38; sample 2, n = 69) were evaluated independently by 2 examiners. In this in vitro study, evaluation of plaque in the Nyvad system was not possible. Histology (sample 2) was used to validate lesion depth. Area under ROC curves (A(z)), sensitivity, specificity and percent agreement of both systems were calculated at D1, D2 and D3 thresholds. Both systems showed kappa for intra- and inter-examiner agreement >0.86 and good correlation with histology: Spearman rho = 0.73 (Nyvad) and 0.78 (ICDAS-II). They presented similar performances except that ICDAS-II showed significantly higher sensitivity (Nyvad 0.89; ICDAS 0.92) and A(z) (Nyvad 0.85; ICDAS 0.90) for the D1 threshold. The correlation between the systems for lesion activity (Cramer's V) was 0.71. Therefore, both visual systems are reliable and can estimate caries lesion depth on primary teeth. Under in vitro conditions, there is no major difference between the Nyvad system and Lesion Activity Assessment in assessing caries activity.