Search is probably the most frequent activity on the Web. Yet it is not effortless,
mainly due to heterogeneous information resources. Semantic search is a means to tackle the
problem of ambiguity. In this paper, we analyse a process of constructing semantic-linguistic
Feature Vectors (FV) used in our semantic search approach. These FVs are built based on
domain semantics encoded in an ontology and enhanced by relevant terminology from Web
documents. Since FVs are central building blocks of the approach, we investigate the quality of
FVs. We take a closer look at the process of FV construction and the impact of chosen
techniques on the quality of FVs. We report on a set of laboratory experiments and analyse
aspects affecting the FV quality and the FV construction error rates.