The arithmetic is trivial. The interpretation is not.
KD ratio is kills divided by deaths. That is the whole calculation. A player with 120 kills and 95 deaths has a KD of 120 ÷ 95 = 1.26.
The number is easy. Deciding whether 1.26 is good is the part that actually needs thought, and the honest answer is that it depends on which game you are playing and what the mode rewards.
Why 1.0 is the real dividing line
In any game where players are matched against each other, every kill is somebody else's death. Across the whole player base, total kills and total deaths are the same number. That makes the average KD exactly 1.0, by definition — not by convention.
So 1.0 is not a modest score. It means you are dead level with the field. Above it, you are winning more duels than you lose.
KDA counts the work KD ignores
KD discards assists entirely, which badly undersells anyone who plays support, uses utility, or does the damage that lets a teammate finish the kill. KDA fixes that by adding assists to the numerator:
(kills + assists) ÷ deaths
Our player with 40 assists goes from a 1.26 KD to a 1.68 KDA. Same performance, and arguably the second number describes it better.
What counts as good, by game type
Respawn shooters — most Call of Duty modes, Battlefield — generate high kill and death counts in every match, and KDs spread wider as a result. Round-based tactical shooters like Counter-Strike and Valorant are different: you get one life per round, so deaths are scarce and expensive. A 1.1 in Valorant reflects a stronger performance than a 1.1 in a respawn playlist.
Battle royales distort it further, because most matches end in a death by design and placement matters more than fragging.
When KD is the wrong number to watch
In objective modes, KD can move in the opposite direction to winning. Planting the spike, holding the hardpoint or carrying the flag all involve dying in places a KD-focused player would avoid. Plenty of high-KD players lose games by never taking the risk that wins them.
Treat it as one reading on the dashboard rather than the score itself.