Falls are one of the prominent causes of injury in elderly. A fall detection could help reduce the health risk following the fall that would otherwise get overlooked. Many research studies mostly focus on distinguish a fall from other activities in daily life using smartphone. However, one major problem is a false positive created by a smartphone drop. In this paper, we propose a 3-phase threshold based fall detection algorithm for smartphone which can distinguish a fall from a smartphone drop. The experimental results show that our algorithm achieves a better performance than 2-phase threshold algorithm. Moreover, in smartphone drop cases, our algorithm has 72% specificity higher than 2-phase threshold algorithm which has 31% specificity.