This paper presents experimental results and finite element analysis of hot upsetting of titanium alloys Ti64 and Ti407 using a dilatometer in loading mode. All samples showed barrelling, as a consequence of an inhomogeneous temperature distribution and friction. The FE analysis is a full thermomechanical model of the test calibrated using multiple thermocouples. At each nominal temperature and strain rate, the true flow stress–strain response is inferred using the difference between the initially assumed constitutive response input to the FE analysis, $$\sigma =f\left(T,\dot{\varepsilon },\varepsilon \right)$$
σ
=
f
T
,
ε
˙
,
ε
, and the predicted response of the model. The analysis applies new procedures for: (1) modeling the thermal gradient; (2) finding the flow stress correction due to the inhomogeneity, using literature data as the input to the FE analysis; and (3) smoothing the constitutive data, fitting empirical $$\sigma =f\left(T,\dot{\varepsilon }\right)$$
σ
=
f
T
,
ε
˙
surfaces at multiple discrete strains. The extracted true constitutive data confirm the moderate strain-softening behavior in Ti64 alloy, and the FE model predicts the distribution of local deformation conditions, for application in interpretation of microstructure and texture evolution. This highlights the difference between nominal and actual test conditions, showing that the discrepancy varies systematically with test conditions, with the central strain and strain rate being magnified significantly, by factors of order 2–3.