The 'DataFurnace' code, based on the simulated annealing algorithm, was used to extract elemental depth profiles from Rutherford backscattering spectra of 4-layer and 21-layer anti-reflection coatings and a 22layer narrow-bandpass filter (respectively 240, 1140 and 800 nm thick). The layers are on a sodium glass containing four major elements and the coatings are respectively titania/silica and zirconia/silica multilayers and a yttria/zinc sulphide/silver multilayer. The DataFurnace code has an execution time that increases exponentially with the number of elements in the sample, so to reduce computation time as well as ambiguity we have extended the code to allow fitting for molecules. Accurate fitting requires a proper accounting for the energy straggling of the beam. Multiple spectra from the same sample can also reduce both computation time and the ambiguity of the solution obtained. No information about the samples has to be given to the code except the elements and molecules present. The precision of the extracted depth profiles can be estimated using Bayesian techniques and is 1.5 nm per layer, on average, for the 1100 nm film. Greater precision is available by using high-resolution techniques or a maximum entropy Bayesian prior (neither is done in this work). Where the depth profiles can be represented by layers of specified thickness (in thin-film units), the accuracy is determined largely by the accuracy of the instrumental sensitivity calibration, which is not subject to matrix effects and can be as good as 1%, although multiple and plural scattering effects must be quantified properly for this to be achieved for thick films (not done in this work). Otherwise, the accuracy of the extracted depth profiles is determined largely by the accuracy with which the energy losses are known.