Using an idealized model it is shown that, to minimize errors in double-spiked isotopic analyses, spike mass numbers should alternate with normal. The comprehensive error analysis which follows assumes that the individual ion currents are the independent variates for statistical purposes: errors in these are obtained by combining (i) a coefficient of variation e<supb>p due to fluctuations in ion emission with (ii)</sup> : the noise indices are the ratios (error due to noise)/(error (i) due to emission fluctuation) for the largest ion beam in the spectrum. Methods of computing error multipliers (operating on e<supb>p) from e_R, eb</sup> and the isotopic abundances, are developed. For a particular element each error multiplier can be regarded as a function of two variables, the ratio of the two spikes selected and the spike/normal ratio, which can then be optimized by numerical methods. Optimum values of spike composition and spike/normal ratio are given for observations of isotopic fractionation of lead, molybdenum and iron using various pairs of spike mass numbers. Some possible deviations from first-order fractionation theory are considered. A general algebraic analysis for the determination of radiogenic isotopes is given, and different approaches to fractionation computations are discussed.