The maximum likelihood (ML) method is presented for censored (below detection limit) regression analysis for nonlinear models. The proposed ML method has been translated into an equivalent method of least squares (ML-LS). An iterative algorithm is proposed in two stages to estimate statistical parameters from least squares derived translation. The algorithm developed is applied to a nonlinear model for the prediction of CO concentration in ambient air in terms of concentrations of respirable particulate matter (RSPM) and NO2. It has been shown that if censored data are ignored or estimated through simplifications such as
(i) the censored data is equal to the limit of detection,
(ii) the censored data is half the difference between the detection limit and the lower limit (eg zero or background level) or
(iii) the censored data are equal to the lower limit, this can cause a significant bias in the estimated parameters.
The developed ML-LS method provided better parameter estimates than any of the simplifications in the censored data.