Phishing is a deceptive social engineering trick that lures online users to disclose personal and confidential information to the fake websites disguising as legitimate ones in electronic communication. Several machine learning methods have been proposed to detect phishing websites. The major limitation of assumption of a peculiar distribution of the data plagues analysis with machine learning (ML) especially when the number of data points are limited in number. This paper applies classical rough sets analysis (CRSA) for detecting phishing websites from a collection of legitimate and fake websites sourced from an open UCI Repository. The work focuses primarily on comparing the classification performance of CRSA with the traditional ML tools. CRSA does not require any distributional assumption for analysis and does not need any user inputs for the parameters in the analysis unlike the traditional ML tools. The comparison revealed no significant difference in the classification performances of the two sets of algorithms. The precision of CRSA was 95.6% which is comparable and in some cases better than the other ML tools. Similarly, high ROC and f-measure places CRSA at par with the statistical ML tools. The relaxation in considering the statistical properties of data clearly gives CRSA edge over other data analysis tools. The CRSA based techniques provide a robust method of detecting phishing websites. The generalization of the analysis makes CRSA a quick-to-use framework for detection of phishing websites without consideration of statistical properties of the data.