The solar magnetic fields observed in active regions and their residues are thought to be parts of toroidal field systems renewed every 11-yr cycle from a poloidal field. The latter may be either a reversing (dynamo) field or a non-reversing, primordial field. The latter view was held for some 70 yr, but the apparent reversals of the polar-cap fields in 1957-8 and the development of dynamo theory brought wide acceptance of the former. Here we consider evidence for and against each model, with these conclusions. (i) Several errors combine so that the non-spot measurements of gross magnetic fluxes are too low by factors of 10 or more. A permanent field of 2 G or more might remain unobserved. (ii) Measurements of' average' magnetic field strength are subject to various large errors. In particular, the reported reversals of the polar-cap fields are better explained in terms of tilts of toroidal field residues. (iii) Observations of new-cycle magnetic fields among old-cycle fields, of the gradual fading away of large unipolar regions, and the ubiquitous jumble of very small magnetic loop structures appear explicable only in terms of a primordial field. (iv) More positive evidence of a primordial field is found in the extreme order, symmetry and long-term stability of the polar cap streamers or rays. During one eclipse (1954) the primordial field was seen in the absence of all toroidal field residues. (v) A form of reversal of the interplanetary magnetic field is re-interpreted and shown to be consistent with a primordial, but not a dynamo, field. (vi) A test for a primordial field is that the fields below coronal holes should tend to be positive (outwards) in the northern hemisphere and negative in the southern hemisphere. (vii) Further evidence may be available by studying various plasma structures below coronal holes. An urgent requirement is a study of fibrils, faculae, macrospicules and rays in these regions.