In the non-invasive blood components measurement using near infrared spectroscopy, the useful signals caused by the concentration variation in the interested components, such as glucose, hemoglobin, albumin etc., are relative weak. Then the signals may be greatly disturbed by a lot of noises in various ways. We improved the signals by using the optimum path-length for the used wavelength to get a maximum variation of transmitted light intensity when the concentration of a component varies. And after the path-length optimization for every wavelength in 1000-2500 nm, we present the detection limits for the components, including glucose, hemoglobin and albumin, when measuring them in a tissue phantom. The evaluated detection limits could be the best reachable precision level since it assumed the measurement uses a high signal-to-noise ratio (SNR) signal and the optimum path-length. From the results, available wavelengths in 1000-2500 nm for the three component measurements can be screened by comparing their detection limit values with their measurement limit requirements. For other blood components measurement, the evaluation their detection limits could also be designed using the method proposed in this paper. Moreover, we use an equation to estimate the absorbance at the optimum path-length for every wavelength in 1000-2500 nm caused by the three components. It could be an easy way to realize the evaluation because adjusting the sample cell’s size to the precise path-length value for every wavelength is not necessary. This equation could also be referred to other blood components measurement using the optimum path-length for every used wavelength.