A quantum phase perceptron (QPP) based on phase representation of quantum states is proposed in this study. The proposed quantum perceptron is constructed by a single-layer structure with two inputs and one output. The input |0〉 and |1〉 qubits are transformed into phase representations with amplitude 1 through H gate phase coding. After quantum coding, all operations of the network are performed phase shift and phase sum separately to obtain a single output of the network, and all operations satisfy unitary operation. The Ry operator that rotates around the y-axis is used to realize the phase shift, and the quantum controlled non-gate (CNOT) is used to realize the quantum phase superposition of two qubits. The probability amplitude of 1 in the output quantum state of the network is measured, and the real output is obtained. This study also deduces the learning algorithm of quantum perceptron network weights to solve the “XOR” problem. The experimental results show that for many initial weight phases, the proposed QPP can realize the XOR and XNOR functions which the classical perceptron cannot solved.