The aim of this paper is to give a brief summary concerning important methodological aspects in establishing the reliability of empirical water quality data. These considerations are relevant for applied work e.g. monitoring programmes, as well as theoretical research, e.g. to validate models. The paper concerns data from Swedish lakes on Hg in pike, perch, water and sediments, and a broad set of limnological data (pH, Secchi depth, temperature, alkalinity, total-P, conductivity, Fe, Ca, hardness, chlorophyll-a and colour). These standard parameters generally vary in a lake, both temporally and areally. The focus of this paper is on such variations and how to express lake-typical values. There are large differences in analytical reliability for different parametres; e.g. Hg (in fish and sediments but not in water) and lake pH can generally be determined with a comparatively great accuracy; the average relative standard deviation (V) is only about 2-3 % for pH. Colour, Fe-, total-P-concentration and alkalinity, on the other hand, generally give high V-values. In natural lakes, the variability is often at least twice as large as the "methodological" variability for parameters such as colour, P, Fe and atkalinity (V-values ranging between 20 and 40 % on average in our lakes). This implies that for most parameters one must analyse many samples to obtain representative, lake-typical values with a given statistical reliability. A general furmula expressing how many samples are required to establish lake-typical mean values is discussed as well as statistical aspects concerning the range of empirical data in models based on such data.
