When 14C signals approach background levels, the validity of assumptions concerning Poisson counting statistics and measurement system stability becomes crucial in interpreting the resultant low-level counting observations. This has been demonstrated in our previous work on detection limits for non-Poisson error and it is critical in our current studies of carbonaceous pollutants, where the 14C signal from just 5 mg C is comparable to that of the background for our miniature gas proportional counters. To assure data quality, our multi-detector system is designed for the on-line monitoring of critical parameters that reflect both the (statistical) nature of the non-Poisson errors and the underlying (physical) causes. It sends >60 bits of information/pulse to a microprocessor which automatically generates, for each counting period, two-dimensional spectra and multiparameter correlation and control charts. To evaluate the validity of long-term counting of 1–10 mg C we use robust (statistical) estimators, optimal counting interval subdivision, and time series analysis of the individual pulses. New opportunities for selective sampling and chemical fractionation which come with the small sample measurement capability have led us to give special attention also to higher control levels, involving e g, isotonic heterogeneity and representative standard materials.