Web forum is an online community within which members with common interest exchange ideas. Large amount of content is being generated on a daily basis on variety of topics. Knowledge discovery from this domain is becoming a hot issue in the research community. In this paper, we explore the lexical and non-lexical approaches for mining answers to questions raised in the web forum with the aim of striking a balance between the two. We empirically address the following questions in the paper. Are lexical features better than non-lexical? Can better performance be realized by combining the two? Would there be a significant variability when different classifiers are used? Are the performances of the features influenced by the domain? Both technical and non-technical forum datasets are used in the study. The experimental results are quite revealing and will serve as a pointer for future exploration in the domain.