For data that can be arranged in populations, multivariate data analysis of digitalized distributions is suggested as an alternative method to detect variations. This approach includes a representation of the data that avoids information destroying pre-processing such as averaging. The method is exemplified and discussed for both single variable distributions and multivariable distributions. A theoretical discussion is presented on its use with data from fibre measuring systems, time series and image analysis data. The method is suggested as either a complement or alternative to other types of data analysis and opens new possibilities for variation detection.