In many retinopathies one of the first alterations which can be detected is tortuosity of the retinal vessels. Already published techniques focused on the representation of the vessel as a mathematical curve, which provided good results but problems due to skeletonization or the choice of the sampling rate were shown recently. We present here a new algorithm for the automated grading of tortuosity, which we have applied to images created from the RET-TORT database. The algorithm is based on a combination of multiscale wavelet and nonlinear derived analysis and can be directly applied to images of segmented vessels without suffering from influences of imperfect mathematical abstractions or poorly chosen sampling rates. This improves reproducibility and is advantageous in clinical practice for identification of tortuosity and treatment decision making. Our method is robust against noise and provides equally good results for arterioles and venules, in line with manual rankings by specialists.