Disclosed is a method of selectively predicting hydrocarbon concentration in a sample of unknown hydrocarbon concentration (the unknown sample). The method comprises the steps of: (i) separately subjecting two or more samples of known hydrocarbon concentration to infrared (IR) radiation; (ii) separately detecting an IR signal from the samples of known hydrocarbon concentration; (iii) analysing the IR signals using a multivariate chemometric technique to produce a training data set; (iv) generating a predictive model for hydrocarbon concentration based on the training data set; (v) subjecting the unknown sample to infrared (IR) radiation; (vi) detecting an IR signal from the unknown sample; (vii) applying the predictive model to the IR signal from the unknown sample; and thereafter (viii) selectively predicting hydrocarbon concentration in the unknown sample. Also disclosed is a method for generating a model to selectively predict hydrocarbon concentration in a sample of unknown hydrocarbon concentration and a software program for selectively predicting hydrocarbon concentration from inputted IR signal data of an unknown sample.