Yes, you can use a t test. The lack of randomization means that you can't interpret any difference you detect as being caused by the grouping variable.

If data are not normally distributed, you may use a nonparametric test, such as Mann Whitney U. Results will be more accurate.