Unless you use a bridge your ISP can determine that you use Tor very easily, if they want to. Your ISP can not see the plaintext of your communications as they are encrypted in multiple layers of AES. They can probably determine to some probability the websites that you are visiting if they run a traffic classifier against the encrypted streams, but the best anyone has done at classifying encrypted Tor traffic is a bit above 50% accuracy (check chaos computer club tor traffic classifier). A classifier that uses hidden markov models might have much better accuracy though, I believe the chaos computer clubs classifier only attempted to identify single encrypted pages rather than entire encrypted websites. Continuously classifying a targets encrypted traffic as they surf through multiple linked pages of a fingerprinted website will probably result in significantly higher accuracy versus trying to determine if a target has loaded single pages on a website as a discrete actions. This is not to say that an attacker will be able to determine the plaintext communications you send through Tor, but it is possible that an attacker who can only observe your entry traffic could determine with high probability that you are surfing a website they have fingerprinted with a traffic classifier. I wouldn't lose much sleep over it though.