OBJECTIVE: To investigate in vivo the correlation between the bone microarchitecture of implant bone sites, bone tissue classification subjectively assessed in radiographs and primary implant stability. MATERIAL AND METHODS: Periapical and panoramic radiographs were obtained from 32 partially edentulous patients. Three surgeons classified bone quality at implant sites using two different methods: assessments in periapical and panoramic radiographs (PP) and according to the classification proposed by Lekholm and Zarb (L&Z). During the implant insertion, bone biopsies were taken, and three-dimensional parameters were measured by microcomputed tomography (microCT). Insertion torque value (ITV) and initial implant stability quotient (ISQ) were recorded at the moment of the implantation. ISQ was also recorded at the uncovering stage of a traditional implant protocol. RESULTS: Bone types 2 and 3 were the most prevalent classifications according to PP (54.3%) and L&Z (58.7%). The mean and standard deviation values of primary stability variables were 38.7(16.7) for ITV, 75.3 (7.7) for initial ISQ and 79.3 (6.8) for uncovering ISQ. Several microCT original and factor variables were found to correlate with bone tissue classifications and primary stability variables. L&Z correlated with architecture (r = 0.31; P < 0.05), density (r = -0.43; P < 0.01) and bulk (r = -0.35; P < 0.05), whereas ITV correlated with architecture (r = -0.40; P < 0.01) and density (r = 0.51; P < 0.01). Multiple linear regression analysis revealed that density and bulk explained 32% of the variability of L&Z bone classification, while density and architecture explained 42% of the variability of ITV. CONCLUSIONS: This study demonstrates the first step in validating L&Z classification. MicroCT provides objective and detailed quantitative data on bone microarchitecture. Intraosseous implant stability is mainly determined by the density of the bone. ITVs could be a good indicator of primary implant stability, whereas ISQs measurements have some limitations and should not be used alone.