It is very difficult, because the influence of experimental procedure on the system leads to a change in its state. However, Steven T. Bramwell investigated artificial spin ice by magnetic force microscopy and determined the structural orientation of the magnetic moments inherent in spin ice. Perhaps this will help: Condensed-matter physics: Great moments in disorder // Nature, 2006, 439, 273 http://www.nature.com/nature/journal/v439/n7074/pdf/439273a.pdf
AC susceptibility measurement, and neutron scattering? Spin glass is disordered but locally correlated phase, while spin-ice has long range non-coplanner AF order. Analysing the results of the above two experiments should give a distrinction...
Both are frustrated magnets, although disorder also characterise spin glass systems. Probably you can differentiate them with AC susceptibility. Both present a frequency dependence of the Chi '' cusp. This is because there are many timescales involved in the freezing proccess, but if you see Mydosh's book the temperature shift in the standard spin glass cusp is much shorter than in spin ice. . This happens at least in DTO and HTO. DTO cusp varies from 400 mK to 10K in a range frequency of mHz to KHz.
Another method is measuring the residual entropy. The reason they are called spin ice is because their analogy with water icer The large degenerate ground state of spin ice produces a residual entropy =R/2 ln (3/2) that agrees with the residual entropy explained by Pauling in water ice.
In fact, spin ice are not antiferromagnet, there are evidences that the interactions are FM. This is another difference between dipolar spin ice and other geometrically frustrated magnet , the ferromagnetic coupling is observed in a positive Curie-Weiss parameter.