Fear conditioning, including variants such as delay and trace conditioning that depend on different neural systems, is widely used to behaviorally characterize genetically altered mice. We present data from three strains of mice, C57/BL6 (C57), 129/SvlmJ (129), and a hybrid strain of the two (F(1) hybrids), trained on various versions of a trace fear-conditioning protocol. The initial version was taken from the literature but included unpaired control groups to assess nonassociative effects on test performance. We observed high levels of nonassociative freezing in both contextual and cued test conditions. In particular, nonassociative freezing in unpaired control groups was equivalent to freezing shown by paired groups in the tests for trace conditioning. A number of pilot studies resulted in a new protocol that yielded strong context conditioning and low levels of nonassociative freezing in all mouse strains. During the trace-CS test in this protocol, freezing in unpaired controls remained low in all strains, and both the C57s and F(1) hybrids showed reliable associative trace fear conditioning. Trace conditioning, however, was not obtained in the 129 mice. Our findings indicate that caution is warranted in interpreting mouse fear-conditioning studies that lack control conditions to address nonassociative effects. They also reveal a final set of parameters that are important for minimizing such nonassociative effects and demonstrate strain differences across performance in mouse contextual and trace fear conditioning.