In cone-beam computed tomography (CT), imprecise knowledge of the acquisition geometry can severely impact the quality of the reconstructed image. This paper investigates geometric calibration using data consistency conditions (DCCs). Unlike the usual marker-based off-line methods, the proposed method does not require the extra-scan of a calibration phantom. It is based on the minimization of a cost function, which measures the inconsistency between pairs of projections. The method has been applied to both simulated and real data. The latter were acquired from a micro-CT system with circular trajectory, for which the problem reduces to identifying global misalignments of the system. When compared to uncorrected reconstruction, the method significantly improved the image quality. When compared to marker-based calibration method, the image quality was similar but no calibration scan was required. Finally, the method can handle axially truncated data. Axial truncation is very common in the medical context but often considered intractable for DCC-based methods. We also demonstrate DCC calibration from real data with axial truncation.