An important issue in topical crawler research is feature extraction, which makes great impact on topic description and page relevance scoring. The existing Odds Ratio method shows high performance on high dimension vectors, whereas it does not work well on low dimension condition. An enhanced method EOR based on Odds Ratio method, with word frequency and distribution rate taken into account, is proposed. The simulation shows a 5% increase on text categorization precision on low and middle feature dimension. Furthermore, by combining EOR score and TF value, namely, TF-EOR to calculate word weight and applying it to topical crawler, 4% increases on both precision and recall are obtained.