Cost functions and cost function gradients for use in radiation treatment planning can be computed based on an approximation of an “isodose” surface. Where a clinical goal is expressed by reference to a threshold isodose surface, a corresponding cost function component can be defined directly by reference to that isodose surface, and a corresponding contribution to the cost function gradient can be approximated by identifying voxels that are intersected by the threshold isodose surface and approximating the gradient of the dose distribution within each such voxel.