Gene expression microarray has become a major source for producing high-throughput experiment data. Data mining has been widely applied to dissect the genetic basis of complex diseases. Mining raw, probe-level data leads to a comprehensive understanding of the overall data set, which is especially useful when the goals of the research are different from the original data producer or contributor. Starting exploration from raw data ensures the integrity of original data from being compromized, thus usually yielding reasonable instinct towards choosing the precise algorithms or techniques for further analysis. In this paper, we present steps towards mining raw microarray data. As a case study of our approach, a public data set related to synchronous and metachronous liver metastatic lesions from colorectal cancer is then used, starting from scratch. The result is verified by previous literature, with more insightful findings discovered.