In the kinematic dynamo problem a fluid motion is specified arbitrarily and the induction equation is solved for non-decaying magnetic fields; it forms part of the larger magnetohydrodynamic (MHD) dynamo problem in which the fluid flow is buoyancy-driven. Although somewhat restrictive, the kinematic problem is important for two reasons: first, it suffers from numerical difficulties that are holding up progress on the MHD problem; secondly, for the geodynamo, it is capable of reproducing details of the observable magnetic field. It is more efficient to study these two aspects for the kinematic dynamo than for the full MHD dynamo. We explore solutions for a family of fluid flows in a sphere, first studied by Kumar & Roberts (1975), that is heuristically representative of convection in a rotating sphere such as the Earth's core. We guard against numerical difficulties by comparing our results with well-understood solutions from the axisymmetric (αω) limit of Braginskii (1964a) and with solutions of the adjoint problem, which must yield identical eigenvalues in an adequate numerical treatment. Previous work has found a range of steady dipolar solutions; here we extend these results and find solutions of other symmetries, notably oscillatory and quadrupolar fields. The surface magnetic fields, important for comparison with observations, have magnetic flux concentrated by downwelling flow. Roberts (1972) found that meridional circulation promoted stationary solutions of the αω-equations, preferred solutions being oscillatory when no such circulation was present. We find analogous results for the full three-dimensional problem, but note that in the latter case the ‘effective’ meridional circulation arising from the non-axisymmetric convection (a concept made precise in the asymptotic limit of Braginskii 1964a) must be considered. Thus stationary solutions are obtained even in the absence of ‘true’ meridional circulation, and the time-dependence can be controlled by the strength of the convection as well as by the meridional circulation. The preference for fields of dipole or quadrupole parity is largely controlled by the sign of the velocity: a reversal of velocity from a case favouring a dipole will favour quadrupole parity, and vice versa. For the comparison problem of Proctor (1977b) this symmetry is exact; for the physical problem the boundary conditions make a difference. The boundary effect is first removed by surrounding the dynamo region with a thick layer of quiescent conducting fluid, and then studied numerically by progressively reducing the thickness of this layer to zero. The insulating boundary contributes to the difficulty of obtaining dynamo action, and to the numerical difficulties encountered. The effect of an inner boundary on dynamo action is also considered, but found to be slight.