Selfie culture has emerged as a ubiquitous instrument for self portrayal in recent years. To portray themselves differently and attractive to others, individuals may risk their life by clicking selfies in dangerous situations. Consequently, selfies have claimed 137 lives around the world since March 2014 until December 2016. In this work, we perform a comprehensiv analysis of the reported selfie-casualties and note various reasons behind these deaths. We perform an in-depth analysis of such selfies posted on social media to identify dangerous selfies and explore a series of statistical models to predict dangerous posts. We find that our multimodal classifier using combination of text-based, image-based and location-based features performs the best in spotting dangerous selfies. Our classifier is trained on 6K annotated selfies collected on Twitter and gives 82% accuracy for identifying whether a selfie posted on Twitter is dangerous or not.