Linear regression analysis is an important practice in earth sciences. Three types of lin-ear relationships between two variables can be identified:(i) very small or no measuring er-rors are associated with the x data and deviations (due to errors of measu rem en t alo n e ) o ccu r only in the y data;(ii)errors of m easurem ent are associated w ith both X and Y data and such errors are the only reason for any deviation ftom the line of best—fit:and (iii) the deviation from the equation is not because of measuring errors in X and Y data alone (i.e.the deviation is real,although the linear correlation is still significant).H ow ever,existing regression m odels are m ainly concerned w ith the first tw o types.H ere,a m ethod has been suggested for regression analysis of the th ird type.S uch a m ethod is better than the traditional OLS(ordi— nary least squares) method in that it is logically consistent and non - biased in charater.