Abstract: Seismic data collected in the field frequently suffer from missing values due to constraints of acquisition cost or terrain. Data reconstruction is a critical step in seismic data preprocessing. Based on the compressed sensing theoretical framework, this study subsampled synthesized data using the 2D random undersampling technique. Then, the 3D seismic data were divided into a series of time slices. By introducing the sparse Shearlet transform and using the convex set projection (POCS) algorithm, this study conducted sequential data reconstruction for various time slices. As a result, a Shearlet transform-based time-domain 3D seismic data reconstruction method was developed. Numerical experiments and measured results demonstrate that the proposed reconstruction method exhibits a higher signal-to-noise ratio, a higher computational speed, and better effects than a Curvelet transform-based approach.