In this paper, we address the problem of searching spoken queries within spoken databases, which is referred to as query-by-example Spoken Term Detection (QbE STD). A knowledge-based posteriorgram representation of speech is proposed. The knowledge of sound pattern of a language can be captured in terms of binary distinctive features (DFs). This idea is tailored for the needs of an STD system. The proposed representation can be used as a front-end of a template-based QbE STD system. Template-based spoken term detection experiments are conducted on TIMIT database. Segmental dynamic time warping (DTW) is used for template matching. The performance of STD system improves from a mean average precision (MAP) score of 68.38% when using multi-layer perceptron (MLP) posteriorgram, to an MAP score of 75.35% when using proposed DF representation.