When a word is presented in the periphery, the letters crowd each other unless they are far apart. Crowding also affects the more-peripheral letters of a centrally-presented word. However, increasing letter spacing does not improve reading rate. We present theoretical and empirical results that resolve this seeming paradox. We measured RSVP reading rates for ordered and unordered text as a function of letter spacing at central and peripheral locations. We show that two assumptions, introduced by Bouma and Legge, predict how reading rate depends on letter size and spacing. In all conditions tested (all spacings and sizes, central and peripheral, ordered and unordered) crowding limits reading.