The amygdala plays a crucial role in most clinical neuroscience models of psychopathy, as it is associated with emotion recognition and regulation. However, many neuroimaging studies, especially the ones with high statistical power, found a null relationship between the amygdala and psychopathy. This challenges the theoretical framework. This lack of consensus may be due to the limitation of focusing on the overall activation of the amygdala rather than investigating the pattern of activity within amygdala. Therefore, voxel-based analysis is needed to investigate the activity pattern within the amygdala and identify potential clusters of neighboring voxels that can predict psychopathy. The current study collected fMRI data from 36 incarcerated women from a North American correction institution with an empathy for pain task. A three-level whole-brain univariate analysis was first conducted to examine the brain activity across groups (high-psychopathy group and control group) and different conditions of empathy for pain task (self-pain, self-no- pain, other-pain, other-no-pain). No significant differences were found in the amygdala, as hypothesized. However, significant outcomes were found in the voxel-based multivariate pattern analysis (MVPA). Specifically, results demonstrate classification accuracy of around 60% for the binary classification between the high-psychopathy and the control groups across all four empathy conditions by leveraging both traditional models (decision tree, random forest, logistic regression) and deep-learning models. The study demonstrates the added value of conducting multivariate pattern analysis within amygdala compared to common univariate analysis.