A technique is described for remote estimation of parameters of submerged or buried objects using a ROV-mounted directive acoustic source and a separately located vertical receiver array. The transmitter emits a train of pulses towards the object, and the scattered echoes are recorded at the receivers for subsequent parameter estimation. Parameters of the object are estimated by nonlinear global minimization of the misfit between the experimentally observed and model-predicted time-series. Two minimization methods, a genetic algorithm (GA) and a differential evolution algorithm (DE), are considered. A fast approximative technique for computing the scattered field, the RK (Ray-Kirchhoff) method, is described and used as forward model at the parameter search. The accuracy of the RK method is assessed in two representative model cases, using an accurate full-field boundary integral equation code (XFEM_S) as reference. Experimental data from a sea trial in the Stockholm archipelago are presented, at which a semi-buried box-shaped object was investigated using a ROV- mounted TOPAS 120 parametric sonar as source. The data are used to estimate seven physical parameters of the object; range, depth, roll, yaw, pitch, density and sound-speed. The estimated parameter values are shown to reduce the model-data misfit significantly compared with those based on prior information only.