The paper presents two new frequentist methods for estimating consistency and inconsistency models in network meta-analysis (NMA), using multivariate random-effects meta-regression. These methods are implemented in standard software packages, such as Stata's mvmeta package. The approach allows for the assessment of whether treatment effect estimates from direct and indirect evidence are consistent, which is crucial for reliable NMA results. Previous methods for NMA have either been limited to two-arm trials or used a Bayesian framework. The proposed methods are computationally efficient, avoid sensitivity to prior choices, and do not rely on Monte Carlo error. They are also two-stage estimation procedures, unlike the one-stage Bayesian approach. The paper illustrates the methods using data on thrombolytic drugs, where many treatment pairs are not directly compared in any trial. The consistency assumption is often questionable, as trials may differ in population or design, leading to potential inconsistency. The paper compares three methods—frequentist consistency and inconsistency models, and a Bayesian approach—using the thrombolytic data. It also discusses how to test for consistency using a global Wald test and how to interpret inconsistency parameters. The results show that the consistency model is supported by the data, with treatments B, E, and G most likely to be the best. The inconsistency model identifies potential sources of inconsistency, such as design differences or indirect comparisons. The paper also discusses the choice of reference categories, heterogeneity models, and the implications of different parameterisations. The methods are applicable to multi-arm trials and can be implemented in various software, including Stata, R, and SAS. The paper concludes that the frequentist approach based on multivariate meta-regression provides a useful and efficient way to analyze NMA data, including the ability to test for consistency and rank treatments.The paper presents two new frequentist methods for estimating consistency and inconsistency models in network meta-analysis (NMA), using multivariate random-effects meta-regression. These methods are implemented in standard software packages, such as Stata's mvmeta package. The approach allows for the assessment of whether treatment effect estimates from direct and indirect evidence are consistent, which is crucial for reliable NMA results. Previous methods for NMA have either been limited to two-arm trials or used a Bayesian framework. The proposed methods are computationally efficient, avoid sensitivity to prior choices, and do not rely on Monte Carlo error. They are also two-stage estimation procedures, unlike the one-stage Bayesian approach. The paper illustrates the methods using data on thrombolytic drugs, where many treatment pairs are not directly compared in any trial. The consistency assumption is often questionable, as trials may differ in population or design, leading to potential inconsistency. The paper compares three methods—frequentist consistency and inconsistency models, and a Bayesian approach—using the thrombolytic data. It also discusses how to test for consistency using a global Wald test and how to interpret inconsistency parameters. The results show that the consistency model is supported by the data, with treatments B, E, and G most likely to be the best. The inconsistency model identifies potential sources of inconsistency, such as design differences or indirect comparisons. The paper also discusses the choice of reference categories, heterogeneity models, and the implications of different parameterisations. The methods are applicable to multi-arm trials and can be implemented in various software, including Stata, R, and SAS. The paper concludes that the frequentist approach based on multivariate meta-regression provides a useful and efficient way to analyze NMA data, including the ability to test for consistency and rank treatments.