Cotton fiber length can be measured by the rapid method of testing fiber beards instead of individual fibers. In this method, only the fiber portion projecting from the fiber clamp can be measured. The length distribution of the projecting portion is very different from that of the original sample. Part 1 of this research reported that the original fiber length distribution of cotton can be modeled by a five-parameter mixed Weibull distribution, and the length distribution of the projecting portion can also be modeled by a five-parameter mixed Weibull distribution with parameters different from the original sample. Based on the results reported in Part 1, this work provides a new approach to estimation of the fiber length distribution of the sample based on the length distribution of the observed projecting portion. The proposed approach is the Partial Least Squares (PLS) regression. The probability density function (PDF) curves from the PLS regression method show a good match with the PDF curves obtained from the experimental data except in the short fiber region. Comparisons of some commonly used length quality parameters between experimental data and PLS regression showed good agreement for UHML and ML. The results indicate the proposed PLS approach for obtaining fiber length distribution from the beard test method is very promising, but additional work is needed to improve the estimation accuracy of short fibers.