Traditionally, cortical neurons have been viewed as specialized for single functions or a few highly related functions. However, at the higher levels of cortical processing, neural specialization waters down in a mix of disparate, seemingly unrelated, information. There is no obvious function that unites the variety of information signaled by individual neurons. We recently showed that in prefrontal cortex these "weird" neurons are a signature of the high dimensionality of the neural representations and they are essential to perform complex cognitive tasks (Rigotti et al. 2013). We performed a similar analysis on the dentate gyrus granule cells (DG GCs). Previous electrophysiological studies (Leutgeb et al., 2007) have indicated DG GCs may encode spatial information through the tuning of their firing fields, where ~30% of DG GCs have been reported to have a single firing field (single field cells, SFCs) while ~60% multiple firing fields (multi-field cells, MFCs). However, it remains unclear whether these tuning properties are stable enough to be used to decode the animal's position. To understand how position is encoded by DG GCs, we used miniaturized head-mounted microscopes to perform functional Calcium imaging of DG GCs as mice foraged in an open field. We show that position can be accurately decoded using population activity in the DG with a precision that is comparable to the animal's body size. This could be achieved both with a linear and a nonlinear decoder using 100 to 600 simultaneously recorded cells. We then identified SFCs, which, similar to that seen in electrophysiological studies, constituted ~1/3rd of the recorded neurons. We decoded position from the SFCs and separately from an equal number of MFCs. The cells used for the comparison are the best in each category and were selected using the Lasso algorithm. The decoding accuracy for the MFCs was comparable or better than that for SFCs. This indicates that multi-field cells, which are more difficult to interpret, can be important for encoding position despite their promiscuous spatial tuning. This multi-field encoding strategy may be similar to that of grid cells, which in theoretical studies has been shown to be more efficient than single field cells in encoding space (Mathis et al. 2012).