A method and a means for retrospectively clustering NMR k-space measurement profiles with respect to both cardiac and respiratory phases were developed to explore strategies for (1) reducing cardiovascular and respiratory flow/motion image artifacts and (2) improving T1 and T2 characterization of the heart. The image data are collected at a uniform rate so that echo (TE) and repetition (TR) times are independent of the varying cardiac cycle R-R interval and/or respiratory motions. Cardiac (C) time, respiratory (R) time or diaphragm position, and NMR data acquisition (A) cycle time are collected by microcomputer in parallel with free running (untriggered) image collection on a standard magnetic resonance imager. After the raw data equivalent of multiple images are collected, the C-A-R phase timing data are uploaded from the microcomputer to the scanner's minicomputer for use in a normalized C-R phase plane clustering of the image raw data. Each profile's position in the C-R phase plane is determined and then clustered into a new set of data, one image being equivalent for each desired C-R phase combination. These raw data are then zero-filled and (optionally) filtered to compensate for the nonuniform k-space sampling and, finally, reconstructed. Cardiac "cines" made from these retrospectively gated images are comparable to similarly phased triggered images. When high time resolution is required, retrospective gating can be expected to show improvements over triggering, especially toward the critical latter part of the cardiac cycle, where coronary artery filling occurs. The system described can readily be assembled from generally available components.