The spectral region around 10 micrometer, showing prominent dust emission bands, is commonly used to derive the chemical composition of protoplanetary dust. Different analysis methods have been proposed for this purpose, but so far, no comparative test has been performed to test their validity. We calculated model spectra of disk models with different geometries and central sources, using a 2D radiative transfer code. These spectra were then fitted in a blind test using four different spectral decomposition methods. We studied the effect of disk structure (flared vs. flat), inclination angle, size of the inner disk hole and stellar luminosity on the fitted chemical composition. Our results show that the derived dust compositions by all methods deviate systematically from the real chemical composition. Out of the four tested spectral decomposition methods, our new two-layer temperature distribution method, differs the least from the input dust composition and the results show the weakest systematic effects. The reason for the deviations of the results given by other methods lies in their simplifying assumptions (e.g. single average grain temperature or one component continuum). We also tested the influence of different noise levels on the results of the spectral decomposition methods. We find that, for ground-based observations (8-13 micrometer) the expected uncertainty in the value of the crystallinity is about 11% for a signal-to-noise ratio of 100, while for space-based observations (7-17 micrometer) the uncertainty is about 5%. On the basis of our results, we propose a recipe for the analysis and interpretation of dust spectroscopy data in the mid-infrared which should be especially valuable for analysing Spitzer data and ground-based infrared spectroscopy data in the 10 micrometer window.