In order to study the lightning attachment process, we have obtained highly resolved (about 100 ns time resolution and about 3.6 m spatial resolution) optical images, electric field measurements, and channel-base current recordings for two dart leader/return-stroke sequences in two lightning flashes triggered using the rocket-and-wire technique at Camp Blanding, Florida. One of these two sequences exhibited an optically discernible upward-propagating discharge that occurred in response to the approaching downward-moving dart leader and connected to this descending leader. This observation provides the first direct evidence of the occurrence of upward connecting discharges in triggered lightning strokes, these strokes being similar to subsequent strokes in natural lightning. The observed upward connecting discharge had a light intensity one order of magnitude lower than its associated downward dart leader, a length of 7--11 m, and a duration of several hundred nanoseconds. The speed of the upward connecting discharge was estimated to be about 2¿107 m/s, which is comparable to that of the downward dart leader. In both dart leader/return-stroke sequences studied, the return stroke was inferred to start at the point of junction between the downward dart leader and the upward connecting discharge and to propagate in both upward and downward directions. This latter inference provides indirect evidence of the occurrence of upward connecting discharges in both dart leader/return-stroke sequences even though one of these sequences did not have a discernible optical image of such a discharge. The length of the upward connecting discharges (observed in one case and inferred from the height of the return-stroke starting point in the other case) is greater for the event that is characterized by the larger leader electric field change and the higher return-stroke peak current. For the two dart leader/return-stroke sequences studied, the upward connecting discharge lengths are estimated to be 7--11 m and 4--7 m, with the corresponding return-stroke peak currents being 21 kA and 12 kA, and the corresponding leader electric field changes 30 m from the rocket launcher being 56 kV/m and 43 kV/m. Additionally, we note that the downward dart leader light pulse generally exhibits little variation in its 10--90% risetime and peak value over some tens of meters above the return-stroke starting point, while the following return-stroke light pulse shows an appreciable increase in risetime and a decrease in peak value while traversing the same section of the lightning channel. Our findings regarding (1) the initially bidirectional development of return-stroke process and (2) the relatively strong attenuation of the upward moving return-stroke light (and by inference current) pulse over the first some tens of meters of the channel may have important implications for return-stroke modeling. ¿ 1999 American Geophysical Union