The previously reported algorithms for deriving line spectra (respondance time distributions) from synthetic or smoothed experimental responses is here extended to experimental data. The earlier algorithm was modified to improve performance in the presence of experimental errors. The effect of smoothing the data with the aid of the cubic spline function was examined. The performance of the modified algorithm was studied comprehensively. Auto-predictions and cross-predictions of storage and loss compliances from the generated line spectra were in excellent agreement. In equally good agreement were the line spectra obtained from compliance data and from stress relaxation data obtained on the same material.