We investigate the potential and accuracy of clustering-based redshift estimation using the method proposed by M��nard et al. (2013). This technique enables the inference of redshift distributions from measurements of the spatial clustering of arbitrary sources, using a set of reference objects for which redshifts are known. We apply it to a sample of spectroscopic galaxies from the Sloan Digital Sky Survey and show that, after carefully controlling the sampling efficiency over the sky, we can estimate redshift distributions with high accuracy. Probing the full colour space of the SDSS galaxies, we show that we can recover the corresponding mean redshifts with an accuracy ranging from $��$z=0.001 to 0.01. We indicate that this mapping can be used to infer the redshift probability distribution of a single galaxy. We show how the lack of information on the galaxy bias limits the accuracy of the inference and show comparisons between clustering redshifts and photometric redshifts for this dataset. This analysis demonstrates, using real data, that clustering-based redshift inference provides a powerful data-driven technique to explore the redshift distribution of arbitrary datasets, without any prior knowledge on the spectral energy distribution of the sources.

13 pages. Submitted to MNRAS. Comments welcome