We classified the ˜10^6 galaxy spectra in SDSS/DR7 (Abazajian et al. 2009) into only 17 major classes (ASK classification; Sánchez Almeida et al. 2010). The algorithm provides the goodness of the classification for each individual spectrum and, therefore, a straightforward way to identify those targets which do not fit in the ASK classes. A significant part of these outliers turn out to be failures of the automatic reduction pipelines. However, a fraction of them represents genuine unusual objects which deserve detailed follow up work to assess their nature. These targets provide a unique opportunity for GTC to carry out serendipitous discoveries. This contribution summarizes the main properties of the outliers.