The present study describes the construction and calibration of a text comprehension item pool for a computer adaptive test (CAT) in elementary school. It investigates the efficiency gain of CATs compared to fixed-item tests (FITs). We administered the item pool to N = 1,975 pupils in the 3^rd (G3) and 4^th (G4) grades to estimate the psychometric properties. The item pool consists of 132 items with confirmed measurement quality. A CAT simulation with 25 content-balanced items demonstrates that the item pool measures text comprehension in G3 and G4 with good reliability (Rel ≈ .92) and an approximately constant standard error. A screening CAT with 8 items achieved good reliability (Rel ≈ .80). Further analyses for the construct validation are reported. A CAT based on the item pool can be up to 19 % more efficient than a comparable FIT and is most applicable for measuring text comprehension in groups with large individual differences and for evaluating learning progress.