Local texture descriptors have outperformed holistic texture descriptors in various pattern recognition applications. However, the local descriptors have limitations that can compromise the data in an image. For instance, the Local Binary Patterns (LBP) are sensitive to noise, Local Ternary Patterns (LTP) use a static threshold for all images making it a challenge to select an optimum threshold for all images in a dataset, and the Local Directional Patterns (LDP) use orientation responses to derive an image gradient disregarding the central pixel and 8−k responses. These limitations lead to the loss of subtle texture features while encoding an image. This paper proposes a Local Directional Ternary Patterns (LDTP) texture descriptor, which not only considers the central pixel in encoding image gradient but also takes into account all directional responses and an adaptive threshold. Findings from empirical records for the MIAS breast cancer dataset and using different classifiers show that LDTP attains a higher accuracy level for both normal/abnormal and benign/malignant classification compared to the other local texture descriptors.