We have developed a novel method for direct determination of the steady-state rates of homogeneous nucleation. The method is applicable to studies of crystallization, aggregation, and similar first-order phase transitions in solutions of proteins or other soluble slow-growing materials with temperature-dependent solubility. Temperature T control of the solution supersaturation allows fast supersaturation changes from a level inductive of nucleation to a level where no nucleation occurs, but existing crystals grow to detectable dimensions. In this way, nucleation takes place only at the first T setting at a constant supersaturation before solution depletion due to crystal growth becomes significant. We use inert oil to cover nucleating solutions and suppress nucleation on the solution-air interface. To obtain reproducible statistical characteristics of the intrinsically random nucleation process, a large number of simultaneous trials take place under identical conditions. First data for the nucleation of the model protein lysozyme under typical crystallization conditions show that protein crystal nucleation may be occurring at or even beyond the boundary of applicability of classical, continuum nucleation models.