Nowadays, with computer supported analysis of databases, constructing bibliometric or scientometric indicators may be considered as easy. In bibliography, many of those indicators may be found. The problem seems to us more to verify the accuracy of the global analysis, including the sampling of data. The global coherence of an analysis depends of the adequacy of all the steps of the analysis. We built and used an experimentation with on-line databases to demonstrate this through an example. Keeping the same protocol for data collections, we will use the same indicators over the various samples.