In this work, the dendritic solidification occuring in 
the $NH_4Cl-H_2O$ system has been explored experimentally 
in a 2D geometry.  These experiments show an order of 
magnitude jump in the growth speed of the dendrites at 
a crossover supersaturation $\pi_{th}$, at which point, 
(``fast'') dendritic growth in the $\langle 111 \rangle$ directions 
becomes sustainable.  At smaller supersaturations, there 
is also a crossover in the preferred growth directions 
from $\langle 100 \rangle$ to $\langle 110 \rangle$.  
This series of crossovers, in conjunction with the chamber 
geometry results in a multitude of interesting growth 
forms and patterns.  
It appears that the growth of the ``fast'' $\langle 111 \rangle$ 
dendrites is for some reason inhibited at supersaturations smaller 
than $\pi_{th}$ although their growth rates may very likely 
exceed those of the $\langle 100 \rangle$ and $\langle 110 \rangle$ 
dendrites.  Furthermore, at supersaturations near and exceeding 
$\pi_{th}$, the growth of the $\langle 100 \rangle$ and 
$\langle 110 \rangle$ dendrites is inhibited.  This results 
in an overall stagnation of growth at supersaturations near 
$\pi_{th}$.  An exponential dependence of $\pi_{th}$ on 
$-\frac{1}{T}$ indicates the effect of a thermally activated 
process (possibly surface diffusion) on the barrier to 
the ``fast'' dendritic growth.
Two mechanisms are proposed to 
contribute to the large jump in speed.  One mechanism 
is that the dendrite body itself grows in the form of a herringbone 
structure, where the herringbones serve as diffusional 
baffles, effectively entraining solute-rich liquid.  The 
other mechanism is a vortical fluid flow at the tip of the 
growing dendrites.  This results from both gravity-driven 
convection and shrinkage-induced fluid flow, in conjuction 
with the chamber geometry.  Both $\pi_{th}$ and $v_{th}$ 
(the velocity of the ``fast'' needles growing at $\pi_{th}$) 
were measured, and a relationship for their product was found, 
which depends on the system parameters.  

In addition to the experimental work, a numerical 
investigation was also carried through, in which both 
a cellular automata (CA) model was developed, as well 
as a pair of coupled partial differential equations (PDE's).  
The CA model proved to be a very useful tool for giving 
insight into the growth of the individual dendrites, whereas 
the PDE's treat a growing pattern of dendrites as a single 
phase.  As a result, the main utility of the PDE's is in 
characterizing the envelope separating the growing dendritic 
pattern from the ``pristine'' liquid.  Both of the models take 
into account predominantly diffusive considerations, and 
therefore do not accurately treat the effect of the vortical flow.  
Despite this, there is extremely good agreement between 
the experimentally observed patterns and those generated by 
the CA model.  This strongly suggests that the patterns 
behind the moving envelope are 
selected based predominantly on diffusive considerations, 
and the effective diffusivity is at most modified by the 
presence of the vortical flow.  Due, however, to the 
preferred relative positions and increased speeds of the 
dendrites, resulting from the vortex interactions, the 
speed and shape of the envelope can not be modelled 
solely on diffusive considerations.  
As a result, it is expected that the PDE's will most accurately 
describe the envelope dynamics in systems, in which vortices 
and fluid flow do not play a strong role.