We have studied a small group of minette dikes and plugs that crop out within a flat-lying sequence of siltstone and shale near Ship Rock, a prominent volcanic throat of tuff breccia in northwestern New Mexico. Seven dikes form a radial pattern about Ship Rock we describe in detail the northeastern dike, which has an outcrop length of about 2,900 m, an average thickness of 2.3 m, and a maximum thickness of 7.2 m. The dike is composed of 35 discrete segments arranged in echelon; orientation. of dike segments ranges systematically from N. 52? E. to N. 66? E. A prominent joint set strikes parallel to the segments and is localized within several tens of meters of the dike. Regional joint patterns display no obvious relation to dike orientation. Small offsets of segment contacts, as well as wedge-shaped bodies of crumpled host rock within segments mark the sites of coalescence of smaller segments during dike growth. Bulges in the dike contact, which represent a nondilational component of growth, indicate that wall rocks were brecciated and eroded during the flow of magma. Breccias make up about 9 percent of the 7,176-m 2 area of the dike, are concentrated in its southwest half, and are commonly associated with its thickest parts. We also describe three subcircular plugs; each plug is smaller than 30 m in diameter, is laterally associated with a dike, and contains abundant breccias. Field evidence indicates that these plugs grew from the dikes by brecciation and erosion of wallrocks and that the bulges in the contact of the northeastern dike represent an initial stage of this process. From continuum-mechanical models of host-rock deformation, we conclude that dike propagation was the dominant mechanism for creating conduits for magma ascent where the host rock was brittle and elastic. At a given driving pressure, dikes dilate to accept greater volumes of magma than plugs, and for a given dilation, less work is done on the host rocks. In addition, the pressure required for dike growth decreases with dike length. From numerical solutions for dilation of cracks oriented like segments of the northeastern dike, we find that we can best model the form of the dike by treating it as composed of 10 cracks rather than 35. We attribute this result to coalescence of adjacent segments below the present outcrop and to inelastic deformation at segment ends. Using a driving pressure of 2 MPa (20 bars), we estimate a shear modulus of about 10^3 MPa for the host rocks, in agreement with laboratory tests on soft shale. A propagation criterion based on stress intensity at the segment ends indicates a fracture toughness of the host rocks of about 100 MPa-m^? , a hundredfold greater than values reported from laboratory tests. Segmentation of fractures is common in many materials and has been observed during fissure eruptions at Kilauea Volcano in Hawaii. At the northeastern dike, we attribute segmentation to local rotation of the direction of least principal compressive stress. From continuum-mechanical models of magma and heat flow in idealized conduits, we conclude that magma flows far more rapidly and with less relative heat loss in plugs than in dikes. Although dikes are the preferred form for emplacement, plugs are the preferred form for the flow of magma. We present a numerical solution for volumetric flow rate and wall heat flux for the northeastern dike and find that although the flow rate is extremely sensitive to conduit geometry, the rate of heat loss to wall rocks is not. During emplacement of the northeastern dike, local flow rate increased where wall rocks were eroded and reached a maximum of about 45 times the mean initial rate, whereas the maximum rate of heat loss to wallrocks increased to only 1.6 times the mean initial rate. An inferred progression from continuous magma flow along a dike to flow from a plug agrees well with observations of volcanic eruptions that begin from fissures and later are localized at discrete vents. We