Influence propagates more widely over social networks, and strongly affects people’s thoughts, opinions, and even behaviors in human life. Influence spread is a more complicated problem that makes the modeling of this process even more challenging. The linear threshold (LT) model provides a simple way to describe the problem. However, this model neglects the decay of individuals’ influence as time lapses. This paper develops a dynamic linear threshold model with decay (DLTD) in which influence decays based on three strategies: (1) Random decay, (2) linear decay, and (3) nonlinear decay. The results on real-world datasets show that compared with the LT model and the independent cascade (IC) model, the DLTD tends to be more accurate especially for the decay with a linear function, and more concentrated to converge to the ground-truth as the size of seed set increases. We also observe that our model is more stable to the networks’ structures. Furthermore, an interesting finding indicates that the balance between community structures’ internal links and external links is more effective for influence spread.