Abstract. Using a semi-analytical model developed by Choudhury and Ferrara (2005) we study the observational constraints on reionization via a principal component analysis (PCA). Assuming that reionization at z > 6 is primarily driven by stellar sources, we decompose the unknown function N_ion(z), representing the number of photons in the IGM per baryon in collapsed objects, into its principal components and constrain the latter using the photoionization rate obtained from Ly-alpha forest Gunn-Peterson optical depth, the WMAP7 electron scattering optical depth and the redshift distribution of Lyman-limit systems at z ∼ 3.5. The main findings of our analysis are: (i) It is sufficient to model N_ion(z) over the redshift range 2 < z < 14 using 5 parameters to extract the maximum information contained within the data, (ii) All quantities related to reionization can be severely constrained for z < 6 because of a large number of data points whereas constraints at z > 6 are relatively loose, (iii) The weak constraints on N_ion(z) at z > 6 do not allow to disentangle different feedback models with present data. There is a clear indication that N_ion(z) must increase at z > 6, thus ruling out reionization by a single stellar population with non-evolving IMF, and/or star-forming efficiency, and/or photon escape fraction. The data allows for non-monotonic N_ion(z) which may contain sharp features around z ∼ 7 and (iv) The PCA implies that reionization must be 99% completed between 5.8 < z < 10.3 (95% confidence level) and is expected to be 50% complete at z &approx 9.5-12. With future data sets, like those obtained by Planck, the z > 6 constraints will be significantly improved.