A comparison of six similarity measures for use in intensity-based 2D-3D medical image registration is presented. The study evaluates the accuracy and robustness of these measures using a "gold-standard" registration calculated with fiducial markers. The measures are tested on a spine phantom with CT scans registered to fluoroscopy images. The results show that the introduction of soft tissue structures and interventional instruments significantly affects the performance of some similarity measures. Two measures, pattern intensity and gradient difference, performed well even in the presence of these structures. The paper compares six similarity measures: normalized cross correlation, entropy of the difference image, mutual information, gradient correlation, pattern intensity, and gradient difference. Each measure is used to compare a fluoroscopic image with a digitally reconstructed radiograph (DRR). The registration is performed within a region of interest containing a single vertebra. The study simulates more clinically realistic fluoroscopy images by overlaying clinical features onto the phantom image. The "gold-standard" registration was calculated using twelve fiducial markers. The registration algorithm uses a multi-resolution approach, starting with a coarse resolution and ending with a fine resolution. The algorithm adjusts the six rigid body degrees of freedom to minimize the similarity measure. The results show that pattern intensity and gradient difference achieved the best performance, with minimal error even when soft tissue and interventional instruments were present. The study concludes that these two measures are most suitable for 2D-3D image registration in medical applications.A comparison of six similarity measures for use in intensity-based 2D-3D medical image registration is presented. The study evaluates the accuracy and robustness of these measures using a "gold-standard" registration calculated with fiducial markers. The measures are tested on a spine phantom with CT scans registered to fluoroscopy images. The results show that the introduction of soft tissue structures and interventional instruments significantly affects the performance of some similarity measures. Two measures, pattern intensity and gradient difference, performed well even in the presence of these structures. The paper compares six similarity measures: normalized cross correlation, entropy of the difference image, mutual information, gradient correlation, pattern intensity, and gradient difference. Each measure is used to compare a fluoroscopic image with a digitally reconstructed radiograph (DRR). The registration is performed within a region of interest containing a single vertebra. The study simulates more clinically realistic fluoroscopy images by overlaying clinical features onto the phantom image. The "gold-standard" registration was calculated using twelve fiducial markers. The registration algorithm uses a multi-resolution approach, starting with a coarse resolution and ending with a fine resolution. The algorithm adjusts the six rigid body degrees of freedom to minimize the similarity measure. The results show that pattern intensity and gradient difference achieved the best performance, with minimal error even when soft tissue and interventional instruments were present. The study concludes that these two measures are most suitable for 2D-3D image registration in medical applications.