OBJECTIVE The authors aimed to compare the efficacy of lumbar vertebral body Hounsfield units (HUs) and pedicle HUs at predicting pedicle screw loosening. METHODS The authors retrospectively assessed 143 patients with L3–5 instrumentation. The patients were classified into one of two groups based on the status of their L3 screws (a screw loosening group or a control group). The pedicle HUs and vertebral HUs of L3 were measured using preoperative lumbar CT scans, and the pedicle HUs were measured in two ways: by excluding or by including cortical bone. RESULTS The screw loosening rate was 20.3% (n = 29/143) at the 12-month follow-up. The vertebral body HUs and pedicle HUs in the screw loosening group were lower than those in the control group (vertebral body group: 98.6 HUs vs 121.4 HUs, p < 0.001; pedicle excluding cortical bone: 208.9 HUs vs 290.5 HUs, p = 0.002; pedicle including cortical bone: 249.4 HUs vs 337.5 HUs, p < 0.001). The pedicle HUs tended to have a higher area under the receiver operating characteristic curve value in predicting screw loosening, compared with that of vertebral body HUs, but the difference was not statistically significant (p > 0.05). Among patients with low vertebral body HUs of ≤ 130, the loosening rate was much lower in patients with pedicle HUs of ≤ 340 than in those with pedicle HUs of > 340 (31.0% vs 13.0%, respectively; p < 0.05). CONCLUSIONS Vertebral body HUs alone are insufficient to accurately evaluate the risk of pedicle screw loosening. Therefore, it is important to collect both the pedicle HU and vertebral body HU measurements for surgical planning.