Web documents in HTML or XML form trees with nodes containing text. A base problem in Web information extraction is to find appropriate queries for informative nodes in trees. We propose to learn queries for nodes in trees automatically from examples. We introduce node selecting tree transducer (NSTT) for representing node queries in trees and show how to induce determinist ic NSTTs in polynomial time from completely annotated examples by methods of grammatical inference. We have implemented learning algorithms for NSTTs, started applying them to Web information extraction, and present first experimental results.