A digital-to-analog converter converts digital data in the floating-point representation into an analog signal with less distortion even when the digital data is small. The mantissa part (d9 TO d0) of the digital data is first converted into a first analog signal by an R-2R resistor ladder network (10). The first analog signal thus obtained is directly supplied to another r-2r resistor ladder network (20) to produce second analog signals whose values are 2<-N> (N = 0, 1, 2, ...) magnifications of the first analog signal. And, one of the second analog signals is selectively outputted in accordance with the exponent part (s2 TO s0) of the digital data. There is also disclosed a digital-to-analog converter which includes a data converter (106 and 107) for converting digital data in the form of a fix-point number into floating point data.