Abstract: Lithology identification, as a major geological task, strongly underpins the exploration of solid minerals, oil, and gas. Since the physical properties of rocks bridge lithologies and geophysical fields, their differences can be used for lithology identification. However, the physical property data of different rocks frequently overlap to some extent, posing challenges to accurate lithology identification using cross plots alone. The K-nearest neighbor (KNN) model is suitable for multi-class classification since it is a simple and direct machine learning method with high accuracy and sensitivity. This study introduced a feature-weighted KNN model for lithology identification. In this model, different weights were assigned to different features by combining the conventional KNN model with the information gain of attribute features. This allowed for intuitive reflection of the importance of attribute features to classification. Experiments show that compared to the conventional KNN model, the feature-weighted KNN model can more significantly identify lithologic boundaries, thus improving the overall accuracy and stability of lithology identification.