Using a proper parametrization, the source displacement field of a seismic event can be efficiently reconstructed by a redundant dictionary of Green’s functions based on sparse representation theory. Then, by subjecting the pre‐existing event records and pre‐computed dictionary of Green’s functions into a sparsity‐promoting algorithm, it is possible to simultaneously evaluate the origin time, hypocentre coordinates and seismic moment tensor. The proposed method is applicable to single‐ or multiple‐source scenarios and, with minor adjustments, can be a valuable tool for real‐time, automatic monitoring systems. This study demonstrates the effectiveness and accuracy of the dictionary‐based approach via (1) detection of microseismic events produced during the hydraulic fracturing of oil and gas wells and (2) inversion of a small‐magnitude, regional earthquake (2002 June 18 in Caborn, Indiana) data. Our experiments based on numerical simulations and earthquake observations underscore the largely untapped potential of dictionary‐based approaches and sparse representation theory in continuous source parameter recovery.