<p>The data-based fault-tolerant bipartite formation problem is studied for heterogeneous multiagent systems under signed graph. The distributed bipartite formation measurement error (DBFME)-based NN is built for estimating actuator fault dynamics, in which information interaction among agents and distributed bipartite formation measurement error are adequately utilized to update neural network weights. Compared with the existing methods, updating weights with neural network estimation is avoided. Furthermore, the proposed DBFME-based NN overcomes the limitations of existing methods that only utilize information from a single agent to estimate fault. By utilizing the acquired fault estimation, a model-free adaptive fault-tolerant bipartite formation control technique is formed based on input/output data of agents, which avoids dependence of existing methods on system mathematical model. The efficacy of built technique is then confirmed using a simulation.</p>