This paper introduces a new algorithm called Boundary-Based Registration (BBR) for accurate and robust brain image alignment. BBR is designed to align images by maximizing the intensity gradient across tissue boundaries, making it more accurate and robust than existing methods like correlation ratio (CR) and normalized mutual information (NMI), especially in cases with strong intensity inhomogeneities or partial brain images. BBR excels at aligning partial brain images to whole brain images, a domain where existing registration algorithms often fail. It is also robust to variations in parameters and initialization, and can accurately register single slices, which is challenging for most current methods. BBR uses a high-quality anatomical reference image to extract surfaces that separate tissue types, and aligns the input image to this reference by maximizing the gradient of intensity across the tissue boundary. Unlike other methods, BBR does not treat the two images as equal, and instead focuses on the contrast across tissue boundaries as the most salient registration cue. The algorithm is based on a cost function derived from the contrast between gray and white matter, and is designed to be robust to spatial intensity inhomogeneities and B0 distortion. The BBR algorithm was evaluated against CR and NMI using blinded human raters and improved fMRI results. It was found to be more accurate and robust, with results that were insensitive to parametric variations in the cost function. BBR was also tested on various scenarios, including B0 masking, reduced field-of-view, intensity inhomogeneity, and inaccurate surfaces, and showed consistent performance across these conditions. The BBR algorithm was implemented as part of the FreeSurfer software package and was tested on a dataset of 18 subjects, each scanned at four sites undergoing both functional and anatomical protocols. The results showed that BBR outperformed CR and NMI in terms of registration accuracy and robustness, with a lower AAD (average absolute distance) between registrations. BBR was also found to be effective in aligning partial brain images and single slices, and was robust to variations in starting points and parameters. The computational load of BBR was also evaluated, and it was found to be efficient, with processing times of up to 15 minutes for 12 DOF registrations on a 64-bit 2GHz XEON processor.This paper introduces a new algorithm called Boundary-Based Registration (BBR) for accurate and robust brain image alignment. BBR is designed to align images by maximizing the intensity gradient across tissue boundaries, making it more accurate and robust than existing methods like correlation ratio (CR) and normalized mutual information (NMI), especially in cases with strong intensity inhomogeneities or partial brain images. BBR excels at aligning partial brain images to whole brain images, a domain where existing registration algorithms often fail. It is also robust to variations in parameters and initialization, and can accurately register single slices, which is challenging for most current methods. BBR uses a high-quality anatomical reference image to extract surfaces that separate tissue types, and aligns the input image to this reference by maximizing the gradient of intensity across the tissue boundary. Unlike other methods, BBR does not treat the two images as equal, and instead focuses on the contrast across tissue boundaries as the most salient registration cue. The algorithm is based on a cost function derived from the contrast between gray and white matter, and is designed to be robust to spatial intensity inhomogeneities and B0 distortion. The BBR algorithm was evaluated against CR and NMI using blinded human raters and improved fMRI results. It was found to be more accurate and robust, with results that were insensitive to parametric variations in the cost function. BBR was also tested on various scenarios, including B0 masking, reduced field-of-view, intensity inhomogeneity, and inaccurate surfaces, and showed consistent performance across these conditions. The BBR algorithm was implemented as part of the FreeSurfer software package and was tested on a dataset of 18 subjects, each scanned at four sites undergoing both functional and anatomical protocols. The results showed that BBR outperformed CR and NMI in terms of registration accuracy and robustness, with a lower AAD (average absolute distance) between registrations. BBR was also found to be effective in aligning partial brain images and single slices, and was robust to variations in starting points and parameters. The computational load of BBR was also evaluated, and it was found to be efficient, with processing times of up to 15 minutes for 12 DOF registrations on a 64-bit 2GHz XEON processor.