A B S T R A C TFor parties of unequal seat shares (s i ), the widely used effective number of parties (N = 1/⌺s i 2 ) offers an equivalent in equal-sized parties, but it needs a supplement to express the imbalance in actual shares. This is akin to supplementing the mean with the standard deviation. A suitable 'index of balance' is b = -log s 1 /log p, where s 1 is the largest share and p is the number of seat-winning parties. It can range from 0 (utter imbalance) to 1 (perfect equality of all parties). In most individual countries, the median balance is between 0.4 and 0.6, and the worldwide median balance is close to 0.50 for any number of seatwinning parties except 2, in line with a simple logical model. Independent of electoral systems used, a rule of conservation emerges: the median product of the largest party's fractional share and the square root of the number of seat-winning parties is conserved: s 1 p 0.5 = 1. The worldwide median for 603 elections is within 2 percent of 1.00.KEY WORDS Ⅲ largest seat share Ⅲ laws of conservation Ⅲ logical quantitative models Ⅲ number of parties When changes in electoral rules or other conditions enable more parties to win seats in a representative assembly, the seat share of the largest party tends to go down. Is there some characteristic of the party system that tends to stay constant in the process? Is something conserved?The concept of a conserved quantity is important in many areas of science. Quantities such as energy, momentum, electric charge and (under certain conditions) mass are conserved when a closed system undergoes changes. It is worthwhile asking whether any quantities tend to be conserved in the course of political processes. Absolute in macroscopic physics, the conservation principles become probabilistic at quantum level. In social relations, a stochastic element can be expected, so that conservation could be expected to apply only to the median outcomes.This study tests a conservation relation for party systems that connects PA R T Y P O L I T I C S V O L 1 1 . N o . 3 pp. 283-298