HVI data produced for commercial purposes is insufficiently precise for use by breeders. Improving the precision of the HVI data is studied on a wide range of saw- or roller-ginned cottons either of commercial origin or aquired from breeding programs. The effects of manually mixing the samples before analysis is tested. Different configurations: MCI-Universal-sampler, MCI-Fibrosampler and ZUS-Fibrosampler are compared. Measurement precision for the different types of cotton is given for each configuration in relation to the number of specimens per sample. Precision is expressed as confidence interval for Upper Half Mean Length, Mean Length, Uniformity Index and Strength. Certain configurations give precision values compatible with breeding work using less than 10 specimens per sample.