Objective: As most studies relating to mental health and disasters have employed cross-sectional or follow-up assessments about psychological health with post-disaster information, the association between changes in social ties and mental health remains unclear. We examined the relationship between the changes in survivor neighborhood ties and depressive symptoms before and after a natural disaster. Methods: Participants were 3567 individuals aged ≥65 years living in Iwanuma city who had responded to questionnaires by the Japan Gerontological Evaluation Study both predating the 2011 Great East Japan Earthquake and Tsunami, and 2.5 years afterward. Changes in the depressive symptoms were assessed using the geriatric depression scale (GDS) at the baseline and follow-up survey. Changes in the neighborhood ties were assessed by asking the participants about their interactions with people in their neighborhood. Possible confounders were adjusted in a linear regression model. Results: Among the 3111 participants in this analysis, 1073 (34.5%) had increased GDS score after the disaster. There were 336 (10.8%) individuals who had neighborhood ties before the disaster, but had no ties afterward; their mean GDS score increased from 2.93 points in 2010 to 3.19 points in 2013. Among those who had not had ties before and after the disaster the mean GDS score remained almost stable, from 2.19 points in 2010 to 2.12 points in 2013. The participants with post-disaster ties were significantly less likely to have an increased GDS score compared with those who had not had ties before and after the disaster (β = -0.39; 95% confidence interval: -0.72, -0.06). Conclusions: Increased neighborhood ties after the disaster reduced the risk of depressive symptoms even when survivors suffered disaster damages. The study reinforces the importance of social capital in disaster recovery and suggests to local governments and local communities that fostering horizontal, neighborhood ties may improve disaster preparedness and mental health resilience.