This study investigates whether insula cortex abnormalities of methamphetamine-dependent subjects (MADs) can detect craving state and predict relapse susceptibility. Voxel based morphometry and statistical parametric mapping were used on structural MRI of MADs. Total 142 MADs were divided into two groups: model-group (n=112) and validation-group (n=30) from follow-up MADs. The results showed that MADs without craving had significantly smaller insula volume. Optimal insula volume determined from Youden index cut-off point on ROC analysis could be used as MRI bio-markers with acceptable accuracy for detecting craving state. Our results could help guide optimally timed intervention, prevention, and treatment strategies for MADs.