One of the stages in oil and gas exploration is a Petrophysical analysis, which aims to determine the structure of rock layers below the earth's surface. The petrophysical analysis uses physical properties in a well-log to determine the rock type below the surface. Nowadays, the software for conducting petrophysical analysis has utilized a machine-learning approach to predict rock types. Most of the software uses the supervised learning method to classify rock types. This research uses a different approach, unsupervised learning, to group rock types based on various features in a well-log. Using a publicly available well-log in Stafford, United States, and the k-means clustering algorithm, this study groups the data into 3 clusters. The result is compared with manual analysis interpretation and shows an alignment between them. From the result, it shows that the unsupervised learning method effectively predicts limestone, shale, and evaporites in the well. It classifies the dataset into useful clusters, generates useful lithologies, provides useful rock characterization, and less time-consuming.