A mathematical method for detecting spectral impurities in liquid chro
matographic peaks measured with a diode array detector (LC-DAD) is des
cribed. The new method compensates for certain measurement non-idealit
ies, minimizing the likelihood of a false indication of an impurity. T
he algorithm, evolving projection analysis, is based on a previously r
eported method in which points in an n-dimensional absorbance space (n
= number of wavelengths) are projected onto two-dimensional subspaces
and modeled using the Kalman filter. Model deviations can be indicati
ve of the presence of spectral impurities, or may result from measurem
ent artifacts such as heteroscedastic noise or non-linear detector res
ponse. The modified algorithm incorporates model variants that mask th
e effects of heteroscedastic noise and non-linearities so that impurit
ies can be detected even in the presence of these artifacts. A noise m
odel is developed for the DAD and simulations are used to evaluate the
modified algorithm. Simulations are validated through the use of expe
rimental data. It is demonstrated that, except in cases where the effe
cts of non-linearities become severe, the modified model successfully
compensates for these measurement non-idealities.