Although ventilatory failure is the most common cause of death in amyotrophic lateral sclerosis (ALS) and measurement of respiratory muscle strength (RMS) has been shown to have prognostic value, no single test of strength can predict the presence of hypercapnia reliably. RMS was measured in 81 ALS patients to evaluate the relationship between tests of RMS and the presence of ventilatory failure, defined as a carbon dioxide tension > or = 6 kPa. We studied the predictive value of vital capacity (VC), static inspiratory and expiratory mouth pressures (MIP, MEP), maximal sniff oesophageal (sniff P(oes)), transdiaphragmatic (sniff P(di)) and nasal (SNP) pressure, cough gastric (cough P(gas)) pressure and transdiaphragmatic pressure after bilateral cervical magnetic phrenic nerve stimulation (CMS P(di)) to identify the risk of ventilatory failure in the whole group and in subgroups of patients with and without significant bulbar involvement. For patients without significant bulbar involvement, sniff P(di) had greatest predictive power [odds ratio (OR) 57] with specificity, sensitivity and positive and negative predictive values (PPV, NPV) of 87, 90, 74 and 95%, respectively Of the less invasive tests, per cent predicted SNP had greater overall predictive power (OR 25, specificity 85%, sensitivity 81%) than per cent predicted VC (9, 89%, 53%) and per cent predicted MIP (6, 83%, 55%). No test had significant predictive power for the presence of hypercapnia when used to measure RMS in a subgroup of patients with significant bulbar weakness. Thirty-five patients underwent polysomnography. CMS P(di), sniff P(di) and per cent predicted SNP were significantly correlated with the apnoea/hypopnoea index (AHI) (P = 0.035, 0.042 and 0.026, respectively). The correlations between AHI and per cent predicted MIP and VC were less strong (both non-significant). In ALS patients without significant bulbar involvement, novel tests of RMS have greater predictive power than conventional tests to predict hypercapnia. In particular, the non-invasive SNP is more sensitive than VC and MIP, suggesting that it could usefully be included in tests of respiratory muscle strength in ALS and will be helpful in assessing the risk of ventilatory failure. In patients with significant bulbar involvement, tests of respiratory muscle strength do not predict hypercapnia. Sleep-disordered breathing is correlated with RMS and the novel tests of RMS having the strongest relationship with the degree of sleep disturbance.