Parallel magnetic resonance imaging (MRI) using multichannel receiver coils has emerged as an effective tool to reduce imaging time in various MRI applications. However, there are still a number of image reconstruction issues that have not been fully addressed, thereby limited the level of speed enhancement achievable with the technology. This paper reformulates the image reconstruction problem as a digital filter bank problem (in contrast to the conventional matrix formulation used in the popular SENSE technique) and develops an effective algorithm to handle the ill-conditioned problem often encountered in image reconstruction. Simulations based on an eightchannel configuration demonstrate that the proposed method is superior to the SENSE reconstruction especially when large acceleration factors are used.