A class of multivariate PDF estimators for use in feature-based target classification in active sonar is presented. The goal is to develop a method that (1) preserves feature identity, (2) captures complex data structures and tails using non-parametric estimates, (3) utilizes parametric models for higher-order inter-feature dependencies, and (4) provides a systematic enhancement to classifier performance given limited target data. The estimators are labeled pseudo-hybrid because (1) they are not true PDFs but they have sensible limiting behavior and (2) they combine non-parametric and parametric models. The simplest form is: PPH(f)=∏i=1Npi(fi)×MN(f)/∏ i=1NMi(fi), where f is the feature set of N features, p’s are the marginal distributions, and MN is a parametric model with marginals Mi. The denominator was chosen so that the overall expression reduces to the correct PDF in the limits that the p’s are independent or MN happens to be the correct model. The efficacy of this PDF estimator to mimic the sampled, test distributions will be demonstrated. Other pseudo-hybrid PDF estimators will also be discussed.