A divergent selection experiment for phytate phosphorus bioavailability (PBA) was undertaken for 3 generations using the Athens-Canadian Randombred (ACRB) chickens. Thirtyfive sires and 105 dams were used to generate the base population (G0). 951 individuals in G0 were measured and were ranked according to their hatch-corrected PBA values to establish the divergent sub-populations. For each line, 12 males and 36 females from individuals with highest or lowest hatch-corrected phenotypic values (or breeding values in high PBA line at G2) were selected as breeders. At generations 1-3 (G1-G3), about 430 individuals per line were measured. PBA was estimated from the disappearance of phytate during the passage of feed through the gastrointestinal tract under a diet containing a suboptimal level of phosphorus in ration. The heritability of PBA and the genetic correlations with body weight (BW), BW gain (BWG), feed intake (FC) and feed conversion ratio (FCR) were estimated with the restricted maximum likelihood (REML) procedure and an animal model. The direct and correlated responses were investigated with fixed effect models and mixed model methodology. Following main results and conclusions were obtained. (1) The inheritance of PBA was not apparently deviated from an additive model of many loci and the heritability was low (0.07 0.02 - 0.09 0.03). The estimates using a pooled data set combining G0 and either high or low PBA line were the same as or close to that with the data of G0 exclusively. The combination of the data sets of the divergent lines led to a lower estimate (0.05 0.02). (2) Selection for PBA proved to be feasible. At G3, the cumulated divergent response (Rc) reached 2.56% (P<0.01). The realized heritability (0.03-0.06) was lower than the estimated values. Best linear unbiased prediction (BLUP) selection was more efficient than individual phenotypic selection. (3) The means of phenotypic values of PBA fluctuated across generations. This implies that, from the short-term selection experiment, it is difficult to establish any trend on the phenotypic level. (4) The least square analysis based on line comparisons detected the divergent selection response but did not indicate the true genetic changes that occurred in each line. The application of mixed model methodology proved to be valid in the separation of observed change into its environmental and genetic components. (5) With the data of the base population (G0) exclusively, the estimated genetic correlations of PBA with BW, BWG and FC were negative and moderate. This was verified by the cumulated divergent correlated responses (DCRC) in these traits. However, the correlated responses were asymmetric, and were mainly attributed to the genetic changes that occurred in the upward selection line. (6) The negative genetic correlations of PBA with BW, BWG and FC became much weaker or even reversed in direction when the information from the selected generations (G1 - G3) were combined with the data of G0 to estimate (co)variance components. There was line dimorphism in the changes.