There are recent interests with CsI(T1) scintillating crystals for Dark Matter experiments. The key merit is the capability to differentiate nuclear recoil (nr) signatures from the background beta/gamma-events due to ambient radioactivity on the basis of their different pulse shapes. One of the major experimental challenges is to perform such pulse shape analysis in the statistics-limited domain where the light output is close to the detection threshold. Using data derived from measurements with low-energy gamma's and nuclear recoils due to neutron elastic scatterings, it was verified that the pulse shapes between beta/gamma-events are different. Several methods of pulse shape discrimination (PSD) are studied, and their relative merits are compared. Full digitization of the pulse shapes is crucial to achieve good discrimination. Advanced software techniques with mean time, neural network and likelihood ratios give rise to satisfactory performance, and are superior to the conventional Double Charge method commonly applied at higher energies. PSD becomes effective starting at a light yield of about 20 photo-electrons. This corresponds to a detection threshold of about 5 keV electron-equivalence energy, or 40-50 keV recoil kinetic energy, in realistic experiments. (C) 2003 Elsevier B.V. All rights reserved.