In this thesis work, three areas that require investigation for transmission of voice/data in a packet network is studied. First, the characteristics of multiplexed voice signals is investigated. It is found that when the number of callers is more than 10, the arrival pattern of talkspurts can be represented approximately by a Poisson distribution. When voice signals are packetized in fixed length, its arrival pattern can also be represented by a Poisson distribution if the number of callers is more than 20. The deviation from a Poisson distribution decreases as the number of caller increases. Second, the performances of two integrated voice/data packet networks of different architectures are compared. In one network, standard link layer protocols intended for data transmission are used for voice packets with no modification. In the other one, voice packets are not processed by those protocols but treated differently from the lowest layer. It is found that the former exhibits more degradation in performance than the latter as the traffic volume becomes large. Lastly, an algorithm that allocates channel capacity to voice and data traffics dynamically to increase the network performance is proposed. In this scheme the maximum delay of voice packets is limited to 200 ms. The improved network performance with this algorithm is shown by simulation. This algorithm incorporated with voice flow control gives much increased network throughput and reduced delay of data packets at the cost of slightly degraded voice quality.