We investigate the mechanism for initiating phreatic eruptions following the emplacement of a shallow magmatic intrusion into water‐saturated permeable rock which contains subsidiary low‐permeability crack networks and disconnected cracks. Heat from the intrusion causes the local groundwater to boil and ascend through the main permeable crack network. As the ascending superheated steam heats the overlying rock, the water in the subsidiary networks and disconnected cracks will boil. The pressure exerted by the vapor in the subsidiary and disconnected cracks can lead to rapid horizontal crack propagation, resulting in an increase in crack length by more than an order of magnitude. According to the model, the eruption process starts near a free surface and migrates rapidly along thermoelastic isostresses as a result of multiple breakage of the thin surface layers above the cracks. For certain crack and rock parameters, however, the crack propagation mechanism, instead of leading to a dynamic eruption, may generate a highly cracked zone that may be removed later by fluid transport processes. The proposed mechanism gives rise to precursory phenomena observed in conjunction with many phreatic eruptions. According to the model developed here, phreatic eruptions are most likely to occur only for a rather restricted set of rock parameters. For example, the country rock should not be too strong (σt ≅ 10 MPa) and should be characterized by two‐scale permeability structure involving a main crack network of relatively high permeability (≳10−12 m2) and a subsidiary crack network with much lower permeability (<10−17 m2). Moreover, the model works better if the mean crack aspect ratio is relatively large (β ∼ 10−1) and the crack concentration is not too low (Ω > 10−2). These restrictions may explain indirectly why phreatic eruptions are not ubiquitous in volcanic regions.