Although electron energy-loss spectroscopy (EELS) in the scanning transmission electron microscope (STEM) provides high sensitivity for measuring the important element, calcium, in biological specimens, the technique has been difficult to apply routinely, because of long acquisition times required. Here we describe a refinement of the complementary analytical technique of energy-filtered transmission electron microscopy (EFTEM), which enables rapid imaging of large cellular regions and measurement of calcium concentrations approaching physiological levels. Extraction of precise quantitative information is possible by averaging large numbers of pixels that are contained in organelles of interest. We employ a modified two-window approach in which the behavior of the background signal in the EELS spectrum can be modeled as a function of specimen thickness t expressed in terms of the inelastic mean free path lambda. By acquiring pairs of images, one above and one below the Ca L(2,3) edge, together with zero-loss and unfiltered images, which are used to determine a relative thickness (t/lambda) map, it is possible to correct the Ca L(2,3) signal for plural scattering. We have evaluated the detection limits of this technique by considering several sources of systematic errors and applied this method to determine mitochondrial total calcium concentrations in freeze-dried cryosections of rapidly frozen stimulated neurons. By analyzing 0.1 microm2 areas of specimen regions that do not contain calcium, it was found that the standard deviation in the measurement of Ca concentrations was about 20 mmol/kg dry weight, corresponding to a Ca:C atomic fraction of approximately 2 x 10(-4). Calcium concentrations in peripheral mitochondria of recently depolarized, and therefore stimulated and Ca loaded, frog sympathetic neurons were in reasonable agreement with previous data.