Abstract. The scaling and squaring method is the most widely used algorithm for computing the exponential of a square matrix A. We introduce an efficient variant that uses a much smaller squaring factor when A 1 and a subdiagonal Padé approximant of low degree, thereby significantly reducing the overall cost and avoiding the potential instability caused by overscaling, while giving forward error of the same magnitude as the standard algorithm. The new algorithm performs well if a rough estimate of the rightmost eigenvalue of A is available and the rightmost eigenvalues do not have widely varying imaginary parts, and it achieves significant speedup over the conventional algorithm especially when A is of large norm. Our algorithm uses the partial fraction form to evaluate the Padé approximant, which makes it suitable for parallelization and directly applicable to computing the action of the matrix exponential exp(A)b, where b is a vector or a tall skinny matrix. For this problem the significantly smaller squaring factor has an even pronounced benefit for efficiency when evaluating the action of the Padé approximant.Key words. matrix exponential, scaling and squaring, subdiagonal Padé approximation, stable matrix, conditioning, matrix function times a vector AMS subject classifications. 15A23, 65F30, 65G501. Introduction. For a given matrix A ∈ C n×n , the matrix exponential e A is among the most important matrix functions, one reason being its close connection to ordinary differential equations; see, e.g., [42, Chapter 8]. We explore the scaling and squaring method [2,23] for its computation, which is the most commonly used in practice, and implemented for example in Matlab's expm. It is based on a type (2 s m, 2 s m) rational approximation to the exponential in a region of the complex plane containing the eigenvalues Λ(A), where s and m are suitably chosen integer parameters. The scaling and squaring method computes e A by first choosing an integer s such that A/2 s has norm of order 1, then taking a rational type (m, m) Padé approximation r(z) to e z , where m is chosen such that e