Methods are described for preparing human erythrocytes for energy dispersive electronprobe X-ray microanalysis in the transmission electron microscope and for the subsequent statistical analysis of data by univariate and multivariate analysis of variance combined with canonical variate analysis. Since no element is independent of others in the chemical make-up of a cell, multielement analysis of all the available data confers upon an individual cell a unique identity. The statistical approach described in this paper allocates an identity to each cell which is then tested for conformity with others of the same sample. Human blood was stored for up to 29 days in acid citrate dextrose solution at +4°C. Monolayers of washed erythrocytes were prepared on carbon-coated, celloidin films on TEM specimen grids, quenched in liquid nitrogen and subsequently freeze-dried. Fifty erythrocytes were analysed at 0, 7, 14, 23 and 29 days and the X-ray emissions (per 100 s) of sodium, phosphorus, sulphur, potassium, iron and a specified background value measured. Chemical data processed by the statistical routine indicated the presence of ‘outliers’—cells which possessed statistically significant differences in their chemical composition from others of the sample. These data were correlated with alterations in erythrocyte shape examined by scanning electron microscopy. Discocytes, prevalent in stored blood on day 0 were characterized by low Na and high P, S, K and Fe. With storage, discocytes were replaced by spheroechinocytes (stages I, II and III) and, later, spherocytes—these cells being present in greatest numbers on days 7, 7–14, 23 and 29, respectively. Individual erythrocytes gained Na and lost K, P, S and Fe during this period. Loss of erythrocyte mass was demonstrated by a reduction in the background radiation.