The problem of working with relative (compositional) data in geochemistry is addressed. Comparison of methods demonstrates that data need to be log-ratio transformed for the data analysis. A hybrid model performs better than a ratio of factors model for mapping the occurrence of felsic intrusions.