This paper compares six similarity measures for 2D-3D medical image registration, focusing on their accuracy and robustness. The measures are evaluated against a "gold-standard" registration calculated using fiducial markers. The study uses a CT scan of a spine phantom registered to a fluoroscopy image, with the registration performed within a user-defined region of interest containing a single vertebra. To simulate more realistic clinical scenarios, additional soft tissue structures and interventional instruments are added to the phantom image. The six similarity measures tested are: 1. Normalized Cross Correlation 2. Entropy of the difference image 3. Mutual Information 4. Gradient Correlation 5. Pattern Intensity 6. Gradient Difference The results show that the introduction of soft tissue structures and interventional instruments significantly affects the performance of some measures. Mutual Information and Cross Correlation are highly sensitive to these features, leading to high failure rates and increased errors. In contrast, Pattern Intensity and Gradient Difference maintain high accuracy and robustness even in the presence of soft tissue and thin line structures. The study concludes that for effective 2D-3D medical image registration, similarity measures must be able to handle the presence of soft tissue and thin line structures. Pattern Intensity and Gradient Difference are identified as the most suitable measures for this purpose.This paper compares six similarity measures for 2D-3D medical image registration, focusing on their accuracy and robustness. The measures are evaluated against a "gold-standard" registration calculated using fiducial markers. The study uses a CT scan of a spine phantom registered to a fluoroscopy image, with the registration performed within a user-defined region of interest containing a single vertebra. To simulate more realistic clinical scenarios, additional soft tissue structures and interventional instruments are added to the phantom image. The six similarity measures tested are: 1. Normalized Cross Correlation 2. Entropy of the difference image 3. Mutual Information 4. Gradient Correlation 5. Pattern Intensity 6. Gradient Difference The results show that the introduction of soft tissue structures and interventional instruments significantly affects the performance of some measures. Mutual Information and Cross Correlation are highly sensitive to these features, leading to high failure rates and increased errors. In contrast, Pattern Intensity and Gradient Difference maintain high accuracy and robustness even in the presence of soft tissue and thin line structures. The study concludes that for effective 2D-3D medical image registration, similarity measures must be able to handle the presence of soft tissue and thin line structures. Pattern Intensity and Gradient Difference are identified as the most suitable measures for this purpose.