A method is for the deconvolution of a statistically noisy spectral dataset is described comprising the steps of: a. obtaining a spectroscopically resolved dataset of measured flux from a sample that has been collected using a suitable detector radiation system; b. generating an initial estimate of the true spectrum; c. modifying the estimate of the true spectrum by a response function of the detector used to collect the measured flux dataset so as to generate an estimate flux dataset; d. computing a merit value for statistical fit between the measured flux dataset and the estimate flux dataset; e. applying a perturbation to a value of the estimate of the true spectrum; f. repeating steps c and d to the estimate of the true spectrum so changed, accepting the change to the estimate of the true spectrum if the resultant merit value indicates an improvement or if the resultant merit value indicates a deterioration of less than a limit margin, and rejecting the change to the estimate of the true spectrum if the resultant merit value indicates a deterioration of more than a limit margin; and g. repeating steps e and f for each further value of the estimate of the true spectrum to obtain a modified estimate of the true spectrum; h. repeating steps c to g for successive modified estimates of the true spectrum while reducing the limit margin. More completely, a method of detection of a spectrally resolved radiation dataset is described embodying the above.