Annealing wet crystals of quartz and berlinite AlPO4 induces water precipitation. The evolution of precipitation has been studied by transmission electron microscopy on samples annealed in various conditions of temperature, pressure and duration. At the beginning one observes unresolved small contrast features, then tiny bubbles connected to sessile dislocation loops become visible. These dislocations are nucleated to relax the fluid pressure in the bubbles. To minimize the corresponding nucleation energy a partial dislocation loop is first nucleated, then a second partial with a complementary Burgers vector rapidly grows in the same plane and joins the first one. By measuring the mean distance X between precipitates for various annealing conditions one deduces the diffusion coefficient D of the water point defects with the help of the relation X/2 = (2D.t)1/2 where t is the duration of the annealing. One finds in the temperature range (350-1000 °C) for quartz D(m2s-1) = 10-12 exp (—95 kJ mole-1/RT). There is no visible effect of the confining pressure, at least between atmospheric pressure and 700 MPa. The α-β transition slightly affects the D values between 500 and 600 °C but except for this temperature range all the other experimental data fit well the above equation. Two theoretical models of water precipitation are developed. They are based on two contrasted hypotheses about the initial mode of water incorporation. In the first case all the water content is assumed to be initially dissolved as a highly supersaturated concentration of substitutional (4H)Si point defects. Water precipitation would thus occur by homogeneous nucleation of critical embryos, this first stage being followed by a stage of growth while new critical embryos would be continuously nucleated. In contrast, in the second case one assumes that the concentration of point defects is initially the equilibrium concentration at the growth conditions which is very low. Almost all the water would thus be incorporated as tiny clusters of water molecules too small to be detected and precipitation would occur by intercluster diffusion, the bigger clusters growing at the expense of the smaller ones which would progressively redissolve. In fact none of these models fully renders account of the experimental results. These results can only be interpreted by an intermediate situation. Water is incorporated during growth as tiny clusters and as a supersaturated concentration of point defects as well. In the quartz studied extensively in this article this concentration of point defects becomes approximately equal to the equilibrium one only at P ⋍ 700 MPa, T ⋍ 550 °C.