Majority of the existing text classification algorithms are based on the “bag of words” (BOW) approach, in which the documents are represented as weighted occurrence frequencies of individual terms. However, semantic relations between terms are ignored in this representation. There are several studies which address this problem by integrating background knowledge such as WordNet, ODP or Wikipedia as a semantic source. However, vast majority of these studies are applied to English texts and to the date there are no similar studies on classification of Turkish documents. We empirically analyze the effect of using Turkish Wikipedia (Vikipedi) as a semantic resource in classification of Turkish documents. Our results demonstrate that performance of classification algorithms can be improved by exploiting Vikipedi concepts. Additionally, we show that Vikipedi concepts have surprisingly large coverage in our datasets which mostly consist of Turkish newspaper articles.