Interconnections between the error in the concentrations, selectivity and limits of detection are shown for multivariate calibration. Definitions derived from error propagation theory were evaluated for inductively coupled plasma optical emission spectrometry (ICP-OES) spectral data. Selectivity values calculated for different definitions were compared. With one possible exception, these definitions of selectivity cannot be used to satisfactorily predict errors in the concentration estimates for multielement and multisensor systems. An approach to estimating limits of overlapped bands that depends on sample composition is presented.