The development of a strategy for the detection of interferences during the routine analysis of unknown samples by graphite furnace atomic absorption spectrometry is discussed. Since indications of interferences must be found in the shape and position of the signal, different methods for the characterisation of the atomic absorption signal are evaluated. For each signal different correlated parameters are monitored, so that the possibility to summarise them in a multivariate test statistic, called Hotelling's T2, is investigated. The use of control limits for this parameter, established from the measurement results in matrix-free solutions, permits the detection of interferences. However, it is much more difficult to say how strongly the quality of the determination is affected by the detected interferences. Another problem that remains to be solved is the fact that it is difficult to compare signals that are measured in different runs.