Abstract: Reservoir recognition and fine description are critical to oil and gas exploration in deep water area. It is rather difficult task since reservoirs are always changing rapidly in lateral, multiphase superimposed vertically and strongly anisotropic. To solve those problems in reservoir prediction and evaluation, this paper adopts high precision spectral imaging and multiphase constraining random geologic modeling as tools to improve the interpretation from qualitative to semi-quantitative and to solve the problems encountered in recognition of deep water sedimentary bodies and reservoir evaluation.