The aim of this multicenter, prospective, randomized cross-over study was to clarify whether on-line conductivity ultrafiltrate kinetic modeling (treatment B), as a substitute for sodium kinetic modeling, is capable of reducing intradialytic cardiovascular instability in comparison with standard treatment (treatment A), by reducing the sodium balance variability. Both treatments were performed by means of a modified hemodiafiltration technique. Treatment A was performed using fixed dialysate conductivity; treatment B made use of the dialysate conductivity derived from a conductivity kinetic model, in order to obtain an end-dialysis ultrafiltrate conductivity at each dialysis session that was equal to the mean value determined in the same patient during the four-week run-in period. Thus, during treatment B, the expected end-dialysis ultrafiltrate conductivity value of each patient should have been constant. The study was carried out according to a multicenter cross-over design of 16 weeks with two treatments (A or B), two sequences (1 = ABB and 2 = BAA), a run-in period of four weeks (period 1, treatment A), and three consecutive experimental periods of four weeks each. Analysis of variance for a cross-over design was used for the statistical analysis. Forty-nine hemodialysis patients prone to intradialytic hypotension (> 25% of sessions) were enrolled from 16 participating centers, and randomly assigned to either sequence 1 (26 patients) or sequence 2 (23 patients). Six patients dropped out and four were protocol violators, which left 39 patients selected for statistical analysis. There was no difference in the average dialysate conductivity, predialysis and end-dialysis plasma water ultrafiltrate conductivity or body weight between treatment A and treatment B. Thus, the observed mean sodium balance was not different and, as expected, only the intra-patient variability of end-dialysis ultrafiltrate conductivity (index of sodium balance variability) was reduced (21%). During treatment A, systolic blood pressure decreased by 23 mm Hg (95% confidence intervals 21 to 24 mm Hg) at the end of dialysis with respect to the pre-dialysis values. Treatment B reduced this intradialytic decrease (P = 0.001) with a maximum effect at the third hour of dialysis (4.4 mm Hg, 95% confidence intervals 1.9 to 6.9 mm Hg, 23% less than during treatment A, P 0.0005) without any period or carry-over effect (P = 0.53 and 0.08, respectively). There was no treatment effect on intradialytic diastolic blood pressure (P = 0.291). In conclusion, intradialytic cardiovascular stability was significantly improved by matching the interdialytic sodium load with intradialytic sodium removal using on-line conductivity ultrafiltrate kinetic modeling as an alternative to sodium kinetic modeling. Although highly significant, this effect was clinically not very large. By applying this conductivity kinetic model to patients with a more variable sodium intake from one session to another, a greater benefit can be expected.