How to use tf.session.run() for testing (not updating network's parameters)?

Normally I use tf.session.run() for training my networks, and use eval() for getting test accuracy or loss. But I see people also use session.run() for getting testing result, which is very strange for me. I thought tf.session.run() is only for training, not testing.

Is there any secret under tf.session.run() that I didn't know?

Thank you very much!