Image reconstruction in diffuse optical tomography (DOT) is a challenging task because its inverse problem is nonlinear, ill-posed, and ill-conditioned. Anatomical guidance from imaging modalities with high spatial resolution can substantially improve the quality of reconstructed DOT images. In this paper, inspired by the kernel methods in machine learning, we proposed a kernel method to introduce anatomical guidance into the DOT image reconstruction. In this kernel method, the optical absorption coefficient at each finite element node is represented as a function of a set of features obtained from anatomical images such as computed tomography (CT) images. Compared with Laplacian approaches that include structural priors, the proposed method does not require image segmentation. The proposed kernel method is validated with numerical simulations of 3D DOT reconstruction using synthetic CT data. 5% Gaussian noise was added to both the numerical DOT measurements and the simulated CT images. The proposed method was also validated by an agar phantom experiment with the anatomical guidance from a cone beam CT scan. The effects of voxel size and number of nearest neighbors in the kernel method on the reconstructed DOT images were studied. The results indicate that the spatial resolution and the accuracy of the reconstructed DOT images have been improved substantially after applying the anatomical guidance with the proposed kernel method comparing to the case without guidance. Furthermore, we demonstrated that the kernel method was able to utilize clinical breast CT images as anatomical guidance without segmentation. In addition, we found that the proposed kernel method was robust to the false positive guidance in the anatomical image.