This paper deals with results of experiments performed with a new version of a free-running 50-MHz RF generator for producing an inductively coupled plasma (ICP) with two gas flows (argon). Samples were introduced as wet aerosols, after nebulization with a cross-flow pneumatic nebulizer. The background at various wavelengths, and the net signals of spectral lines of widely different character were studied as functions of the power input to the ICP (0.8–1.9 kW). The results were linked with those of interference measurements with alkali matrices, and it was found that with the new facilities also, a relatively low power input (∼1.1 kW) combined with a 15 mm observation height and carrier gas flow rates in the range 1.3–1.6 l/min provide the best “compromise conditions” for simultaneous multielement analysis including trace analysis. Intensity ratios of ionic and atomic lines of various elements, and relative intensities of ionic lines of Mg and Al were determined under the stated “compromise conditions”. These results were connected with a measured effective temperature of 5850 K and a literature value of 10 16cm −3 for the electron number density. On this basis, it was made plausible that the substantial departures from LTE found by other authors in rigorous experiments using radial resolution do also manifest themselves under analytical conditions and that a non-LTE mechanism must be the principal ground for some of the favourable analytical properties of argon ICP's, viz. the high sensitivity of ionic lines and the smallness of ionization interferences. A critical examination of the literature including that dealing with a special type d.c. plasma-jet, which shows similarities with a toroidal ICP, and the results found in this work led the authors to a working hypothesis in which the non-LTE mechanism of argon ICP's is attributed to an overpopulation of metastable argon levels and a dual role of metastable argon, that of an “ionizer” (Penning ionization) and that of an “ionizant” (easily ionizable constituent). To picture this mechanism, the authors set up a simple mathematical model, in which an LTE system is perturbed by an inflow of metastable argon atoms. It is shown that this model can qualitatively account for various features and trends, but it is at the same time pointed out which gaps in our knowledge of ICP's must be essentially filled by further experiments to give the model more substance than that of a working hypothesis.