In this paper we present a new approach to volume rendering in which curves of constant intensity in image space, or iso-contours, are computed directly for each view. The generated iso-contours can be used to drive various visualization and feature-detection algorithms. The approach imposes no restriction on the organization of the data points and can accommodate a large class of radially-symmetric filter functions. The technique works well for both perspective and orthographic viewing projections. Each iso-contour is defined by an ordinary differential equation, which is solved numerically using a predictor-corrector method. A key element of the algorithm is the use of image intensity gradients, which we compute by means of a closed-form expression that holds at every point in the image plane. A caching algorithm is described that dramatically accelerates the gradient computations on large datasets. The algorithm is demonstrated on emission-only datasets. We conclude by describing a number of possible enhancements.