The stability of food agglomerates is very important for keeping optimal instant properties as well as flow characteristics. Compression tests have been proven not only to be useful tools in char acterizing attrition, but also excellent descriptors for powder flowability. The purpose of this work was to study the effects of particle size and water activity (a w) on the compression characteristics of selected agglomerated food powders, and then to identify suitable mathematical models by using a non-linear regression program for predicting the compression characteristics of food agglomerates when partial attrition takes place. Three agglomerated food powders - non-fat milk, low fat milk and instant coffee - were classified by size into five or six fractions with a set of RX-29 sieve screens. Each fraction was conditioned at three a w levels, placed in a cylindrical compression cell, and compressed with a piston attached to the crosshead of a TA-XT2 texture analyser. The crosshead speed was 1 mm/s in all tests and the maximum force applied was 245 N. Particle size was found to play a significant role in compression tests in that the greater the particle size, the greater the volume reduction. It was easier to compress the low a w samples, but in all tests changing a w did not significantly affect compression characteristics. Sone's two- parameter model accurately described the combination of compaction and attrition when compres sion pressure did not exceed a certain level, while Peleg's double-exponential model with four parameters best fitted the compression data.