Tissue oximetry parameter detection based on near-infrared spectroscopy has the advantages of being rapid, real-time, and non-invasive, and has been widely used in clinics with important application values. Near-infrared spectroscopic tissue oximetry requires a combination of different wavelengths of near-infrared light to calculate the concentration of oxyhaemoglobin and reduced haemoglobin, but the combination of wavelengths currently used to detect oximetry information is not uniform. The wavelength selection of the light source is directly related to the accuracy of the extraction of tissue blood oxygenation parameters. In order to improve the measurement accuracy, a wavelength selection method based on condition number was established in this study and experimentally verified by tissue mimicry and in vivo experiments. The experimental results show that when the absorption and scattering coefficients of tissue mimics are changed, the error of total haemoglobin concentration calculated by the smallest conditioned number group is the smallest, and the error calculated by the largest conditioned number group is the largest. In the in vivo experiment, when the collected spectral signals were denoised, the smallest conditioned number group calculated oxygen saturation with small error and the largest conditioned number group inverted oxygen saturation with the largest error, which is in accordance with the underlying theory of conditioned numbers. Therefore, it is feasible to optimize and select the wavelength using the conditional number method, which improves the tolerance of error in the calculation of optical parameters into physiological parameters, and provides a methodological basis for the rational selection of wavelength for tissue blood oxygen detection.