Rapid and seemingly effortless word recognition is a virtually unquestioned characteristic of skilled reading, yet the definition and operationalization of the concept of cognitive effort have proven elusive. We investigated the cognitive effort involved in oral and silent word reading using pupillometry among adults (Experiment 1, N = 30; Experiment 2, N = 20) and fourth through sixth graders (Experiment 3, N = 30; Experiment 4, N = 18). We compared multiple pupillary measures (mean, peak, and peak latency) for reading familiar words (real words) and unfamiliar letter strings (pseudowords) varying in length. Converging with the behavioral data for accuracy and response times, pupillary responses demonstrated a greater degree of cognitive effort for pseudowords compared with real words and stronger length effects for pseudowords than for real words. These findings open up new possibilities for studying the issue of effort and effortlessness in the field of word recognition and other fields of skill learning.