With increasing mining depths and excavation volumes comes not only increased rock stresses and more difficult underground mining conditions, but also increased surface effects, in particular from cave mining. The surface effects of deep sublevel cave mining are not well understood and are further explored in this paper, through a case study of the LKAB Kiirunavaara Mine. Two different numerical modelling approaches were used to quantify potential surface effects. The first approach was applied to Sjömalmen (Lake Orebody). This is a non-daylighting portion in the northern end of the mineralisation, above which surface cratering has developed. Three-dimensional (3D) numerical modelling, using the Itasca caving algorithm, was applied to study future mining of Sjömalmen down to Level 1365 m. In the second approach, 2D modelling of the main portion of the Kiirunavaara orebody was conducted, using a caving simulation scheme initially developed at the Luleå University of Technology. This model enabled simulating caving to large depths, in this particular case down to Level 1800 m, for prediction on hangingwall deformations. The actual caving is simulated implicitly in these continuum models. Observational data on cave development and surface cratering, as well as measured ground surface deformations, were used to calibrate the numerical models. For both approaches, deeper mining was shown to significantly affect the ground surface. Ground deformations are not arrested by bulking and/or increased confinement as mining goes deeper. Both modelling approaches have distinct pros and cons. The 2D approach is only applicable to the main portion of the orebody, where 2D geometrical conditions can be reasonably assumed, but calculation times are faster compared to the 3D approach. The models were fairly sensitive to the geomechanical properties and choice of constitutive model. This facilitated calibration, but also implies that an improved characterisation of the rock mass in the cap rock and hangingwall is important for increased reliability in predictive analyses.