Subsidence due to fluid withdrawal from the subsurface is being experienced in many regions. Water extraction for agricultural and civil needs is the most common cause of this phenomenon. However, hydrocarbon production from deep formations sometimes has been blamed for substantial rates of subsidence (e.g., Wilmington, California; Ekofisk, the North Sea). Hydrocarbon reservoirs in sensitive areas (lowlands), such as the Netherlands and the eastern Po River Plain in Italy, which includes the city of Venice, must be maintained under control to avoid actual or alleged subsidence. Because compaction in deep sediments is a relatively new research field, applicable data are sparse, and much work remains to be done. In particular, a strong discrepancy exists between rock compressibility values measured in the laboratory and in the field, these latter ones being probably more representative. Laboratory compressibility values are at least one order of magnitude larger than in situ values. Given this discrepancy, and in order to accommodate also for the variability of natural materials, a stochastic risk analysis is the most reliable tool for predicting subsidence due to hydrocarbon extraction. This approach is here applied to a planned gas production project from offshore fields in the northern Adriatic Sea, in order to assess the risk of subsidence on the coast and the nearby city of Venice. No stochastic theory of subsidence is available to date; consequently, a Monte Carlo technique is applied with respect to soil compressibility, which affects both the reservoir and subsidence simulations. A semi-analytical three-dimensional model based on the theory of Geertsma is used as a subsidence model. The results show that no actual risk of subsidence exists on the coast. Against intuition, the worst case scenarios coincide with the lowest compressibilities, since these values contribute to higher pressure declines in the reservoirs and the adjacent aquifers, and to a faster propagation of pressure change towards the shoreline. The results for the worst-case scenarios are confirmed by more sophisticated non linear, three-dimensional, finite element simulations.