Joint toxicity of Pb2+, Cr3+, Co2+ and Ni2+ toward Photobacterium phosphoreum (Ph. phosphoreum) at the no observed effect concentration (NOEC) was determined through a factorial experiment. A neural network model was designed according to experimental results and employed for toxicity prediction of unary, binary, ternary and quaternary combinations. The mechanism and trends of joint toxicity were interpreted by quantitative structure-activity relationship, Michaelis-Menten kinetic model, and concentration addition (CA) theory. Toxicity was directly related to the covalent index (Xm2r), covalent binding reaction presented a first and zero order reaction at low and high concentration, respectively, and CA accurately predicted toxicity. Additionally, the results showed that low concentrations of heavy metals should be considered when conducting environment risk assessment.