We studied the anisotropic magnetic response of the internal superconductin
g phases of UPt3 and its anisotropic magnetic susceptibility with a capacit
ive torque meter which is very sensitive in high fields. Experiments were p
erformed at temperatures down to 20 mK and at various angles between the c
axis (hexagonal structure) and (B) over right arrow, ranging from 16 degree
s to 82 degrees. The samples were four single crystals grown with different
methods and subjected to different annealing procedures. The normal state
susceptibility has a maximum around 20 K for (B) over right arrow in the a-
b plane which we followed up to 14 T. It may arise from hybridized uranium
ion states split by the hexagonal crystal field. The magnetization curves i
n the superconducting (SC) regime show strong irreversibilities which are h
ighly sample dependent. They are not correlated with the internal SC phase
lines but continue up to a line of fields that lies parallel to the B-c2 cu
rve and even follows its kink at the tetracritical point (T*,B*). In the cl
eanest sample this line is shifted to fields well below the B-C internal ph
ase line which then manifests itself in a pronounced kink of the magnetizat
ion curve indicating an enhanced Ginzburg-Landau parameter kappa. In anothe
r sample the B-C phase line between two of the three internal SC states cou
ld be detected even in the hysteretic region. The enhanced Ginzburg-Landau
parameter kappa means a larger penetration depth and/or a shorter coherence
length, clear evidence for the unconventional character of the B-C phase t
ransition. With our cleanest sample we also observe an anomalous peak effec
t, a region of enhanced flux pinning near B-c2, which is probably related t
o the Fulde-Ferrell-Larkin-Ovchinnikov state. In yet another sample we find
a crossing of the up-down magnetization curves, also near B-c2, but with r
eversed orientation of the magnetization loops. We interpret this in terms
of different flux pinning in the two main crystal directions, possibly in r
elation to the peak effect which is, however, masked in this sample by stro
ng irreversibilities.