The potential applicability of a recurrent neural network (RNN) in the reconstruction of spectra from truncated FIDs was explored. A RNN was trained on a set of simulated full FIDs with varying metabolite concentrations. Then, the performance of the trained RNN was tested on severely truncated FIDs (~95% truncation). Our preliminary study suggests that RNNs may be used in the restoration of truncated FIDs and thus reconstruction of spectra including tiny multiplets. A well trained RNN may be applicable to the situations where data sampling is highly limited such as in cardiac MRS and spectroscopic magnetic resonance fingerprinting (sMRF).