Extrinsic Fabry-Pérot interferometer sensors are widely applied in aerospace, biomedical, electric power, and hydroacoustic detection fields due to their high sensitivity, rapid response, electromagnetic interference resistance, compact size, and low cost. Common demodulation methods for fiber optic EFPI sensors in engineering applications include intensity demodulation, spectral demodulation, correlation, phase-generated carrier, and three-wavelength orthogonal phase demodulation. The three-wavelength demodulation method, characterized by fast demodulation speed, large range, strong anti-interference capability, simple principle, and easy miniaturization, has gained increasing attention and research in recent years. This technology uses three different wavelengths of interference light signals for phase extraction in EFPI sensors. However, imbalances in the three light signals due to various inconsistencies during signal transmission, such as fiber bending loss, optical filter loss, light source flatness, sensor spectral envelope flatness, and photoelectric conversion gain, can lead to signal distortion and significant errors in demodulation results. Amplitude normalization can be applied for calibration of three-wavelength interference signals with phase changes exceeding 2π, but calibration is challenging when sensor phase changes are less than 2π. Particularly in high-frequency signal testing, the sensor's resonance frequency must be much higher than the measured signal, leading to lower sensitivity and phase changes far less than one phase cycle. For small signals less than 0.2π, the decrease in signal-to-noise ratio makes calibration even more difficult. Building on our research group's study of three-wavelength dynamic demodulation algorithms, this paper proposes a three-wavelength small-signal calibration method based on Lissajous curve fitting. By systematically enumerating calibration coefficients and adjusting the light intensity of three-wavelength interference signals, the method evaluates the orthogonal fitting assessment of the Lissajous curves for the two arctangent intermediate variables in the three-wavelength demodulation algorithm. It judges the contour deviation and data point density under different calibration coefficients, ensuring the curve fits a circle centered at the origin for accurate small-signal calibration. A demodulation system was established based on the dynamic signal calibration and demodulation needs of fiber optic EFPI sensors. Three equal-power distributed feedback lasers with central wavelengths of 1 553.33 nm, 1 557.36 nm, and 1 561.42 nm were used as the light sources. The three different wavelengths of light signals completed interference modulation in the fiber optic EFPI sensor through a three-channel dense wavelength division multiplexing optical filter and coupler. The reflected light signals, after filtering through the optical filters, were collected by the three-wavelength demodulation board card using photodiodes and analog-to-digital converter modules, converting the three different wavelengths of interference light signals into digital signals for computation in the upper computer system. Simulation results show that, based on the conditions of the three-wavelength demodulation algorithm, this calibration method can stably obtain calibration coefficients for phase changes within 0.2π for small signals, with a maximum amplitude error not exceeding 1.1% and strong noise resistance. Experimental tests indicate that the maximum error in sensor sensitivity after calibration does not exceed 0.15%, and the maximum nonlinear error of the fitted curve does not exceed 0.6%. This calibration method achieves accurate calibration for small signals in the three-wavelength phase demodulation algorithm, enhancing the range and precision of the three-wavelength algorithm.