In the mid-frequency (MF) range, the number of modes of warships is very large. Predicting power radiated by the underwater part of the shell becomes difficult with a classical modal synthesis. Statistical methods such as SEA (Statistical Energy Analysis) are also limited in their capability of modelling the structural complexity. Virtual SEA method is introduced to get benefit of both deterministic modal synthesis and SEA approaches to build a fully predictive model in the MF range. Virtual SEA analysis works on an original in-vacuo finite element model (FEM) of the warship. The structural model is identified in the MF range to a numerical SEA model of which parameters are extracted from FEM. First the dynamics of the warship structure is reduced to a statistical FRF (Frequency response function) matrix (between sets of observation nodes). The partitioning into subsystems is computed from this FRF matrix by an attractive algorithm which re-orders nodes to obtain weak coupling between subsystems. Then the FRF matrix computed between all nodes is compressed following the portioning scheme. Finally, using inverse method, the various SEA parameters of the subsystems (modal density, mass and coupling loss factors) are identified from previous compressed FRF. The parametric SEA model provided by this process encapsulates the FEM dynamics in the MF range. This SEA model is then analytically coupled to external fluid to predict the power radiated in water. At the end the virtual SEA model can predict power radiated in fluid under mechanical loading in band-frequency format.