We present an initial utility study of a distributional model of verb selectional preferences for 3rd person pronoun resolution in German. We investigate cases in which 3rd person pronouns occur as subjects of transitive verbs. In each such case, the likelihood of inserting one of the antecedent candidates is calculated as the conditional probability of the antecedent candidate given either the verb governing the pronoun or the object of the verb. These probabilities are estimated using a matrix derived from frequency counts in a large corpus. Non-negative matrix factorisation is applied as a sort of semantic smoothing to address the sparsity issue inherent in the approach.