Vulnerability to presentation attacks (PAs) remains one of the main security concerns of the widely used fingerprint-based authentication systems, especially for unattended and remote applications. PAs can be carried out by presenting artifact, corpse, or conformant samples to a biometric sensor with the intention of circumventing the system policy. In this study, we develop a multilayer biometric authentication system robust against PAs by the fusion of fingerprint and heart-signal. In the first layer, artifact attacks are prevented by using a fine-tuned convolutional neural network (CNN). In the second layer, a lightweight CNN is used for the prevention of corpse attacks by using heart-signal (also known as ECG signal) with duration of 0.5 s. In the subsequent layers, robust fingerprint matcher at a specific threshold are utilized for the prevention of conformant attacks. In the final layer, a score-level fusion of the fingerprint and heart-signal is used for biometric authentication. The proposed system was evaluated by different authentication and attack scenarios using a multimodal dataset comprising two public databases of fingerprints and heart-signals available online. The experimental results yielded a false match rate (FMR) of approximately zero (0.1%) with an acceptable false non-match rate (FNMR). The obtained results are encouraging for the incorporation of the system into applications requiring high-security authentication.