Maximum likelihood reconstruction of activity and attenuation (MLAA) from emission data only suffered from the inherent cross-talk between the estimated attenuation and activity distributions. The authors proposed an improved MLAA algorithm by utilising tissue prior atlas (TPA) and a Gibbs prior as prior knowledge. TPA determines the plausible region for each of the typical attenuation coefficients; hence, it imposes statistical condition as a supplement for the exclusive magnetic resonance (MR) information on the reconstruction process of attenuation map. Therefore, along with the soft tissue distribution provided by the segmentation of MR images, an air mask and a bone probability map breakdown the MR low-signal class into four subclasses in order to favour recognition of air and bone. Estimations on attenuation coefficients are realised as a mixture of pseudo-Gaussian distributions. The proposed algorithm is evaluated using the simulated 3D emission data. The proposed MLAA-TPA algorithm is compared with the MR-MLAA algorithm proposed by Heußer et al. Their results demonstrate that the MR-MLAA algorithm performance depends heavily on the MR segmentation accuracy well handled by MLAA-TPA. The quantification results illustrated that the MLAA-TPA outperformed the MR-MLAA algorithm owing to reduction of misclassification and more precise tissue detection.