Abstract: To address the problems of current adaptive sampling approaches,we present a parallel multidimensional adaptive sampling method.Firstly,the multidimensional space is coarsely sampled and adaptively split into subspaces.Secondly,the borders of each subspace are extended and the number of samples is assigned to each subspace according to its size.Thirdly,a KD-tree is adaptively constructed in parallel for subspace.Finally,the subspaces are reconstructed and combined into an entire image by leveraging the gradient values of sampling points.The experimental results demonstrated that our algorithm takes less memory,runs faster than existing multidimensional adaptive sampling methods,and can generate images rapidly with satisfying effects of motion blur,depth-of-field and soft shadows.