Ethereum Gas Fees by Transaction Type

Gas limit isn't a fee — it's a ceiling

Every Ethereum transaction has two gas numbers attached: gas limit (the maximum you authorize) and gas used (what actually ran). The fee you pay is gas_used × gas_price, not gas_limit × gas_price. The limit is a circuit-breaker.

For a plain ETH transfer, this distinction is trivial — 21,000 is both the protocol minimum (intrinsic gas) and the exact usage. The transaction either runs all 21,000 or it doesn't run at all. There's no padding, no refund.

For everything else, gas_used is a function of the contract's code path through your specific input. A Uniswap swap that crosses three liquidity ticks costs more than one that crosses two. A token transfer to a fresh address (no existing balance entry) costs ~25% more than a transfer to a recipient who already holds the token. An NFT mint that triggers a redeem hook costs more than one without.

The numbers in the table are medians, not maxes — your transaction may use 10-30% more or less than shown. If it uses less, the difference is refunded back to you in the same block. If it uses more than your gas limit, the EVM reverts state but still charges you for the burned gas. That's the “out of gas” failure mode and the most expensive way to fail a transaction.

How wallets actually quote you

When MetaMask or Rabby shows you a fee estimate, it's doing four things behind the scenes:

1. Estimate gas usage: calls eth_estimateGas on the target contract with your exact input. This runs the transaction in a sandbox and returns the actual gas_used.
2. Pad it: wallets add a 10-30% safety margin (MetaMask: ~10%, Rabby: ~25%) so the transaction has headroom if state shifts between estimation and inclusion.
3. Pick a gas price: suggests three tiers — “market”, “aggressive”, and a custom option. Aggressive ≈ what miners need to include you in the next 1-2 blocks.
4. Display the maximum: shows you padded_gas × gas_price × ETH/USD. That's why the wallet number is often 15-30% above the live cost in the table above.

Practical implication: don't pay attention to the “max fee” the wallet shows. Look at the confirmed-transaction fee on Etherscan after the fact — that's your actual cost. For non-urgent transactions, switch the wallet's tier from “market” (its default) to “low” to save 30-50% on the gas-price component.

When gas estimation fails

Sometimes your wallet shows “this transaction will likely fail” and refuses to give you a fee estimate. The instinct is to bump the gas limit and try anyway. Don't. The transaction is probably broken upstream — paying for a transaction that will revert just burns money. Three common causes:

• State changed since you opened the dApp. You quoted a swap when liquidity was deep, then a big trader moved the price 0.5% past your slippage tolerance. The estimator simulates with current state and sees a revert. Fix: refresh the dApp, re-quote.
• Approval is missing or stale. You upgraded the contract but didn't re-approve. Or you have a partial allowance and the swap exceeds it. Fix: check the approval; re-approve if needed.
• Token has transfer hooks. Some tokens enforce permissioned transfer logic (allowlists, fees, pause states). The gas estimator can't always model these — it sees a revert in simulation that wouldn't happen in production, or vice versa. Fix: send a small test transaction first.

If you genuinely need to override and submit anyway — for example, the dApp is broken but you know the underlying call is valid — set the gas limit manually to the table value above with a 30% buffer. But that's a last resort. The default response to “will likely fail” is: don't submit.

Same transaction on an L2 — 5-50x cheaper

Every action in the table above runs on Arbitrum, Optimism, Base, and Polygon with the same contract interface (the major dApps deploy across all four) but at a fraction of the mainnet cost. A swap that costs $0.034 on mainnet is typically $0.05-0.50 on Arbitrum or Base. The same Uniswap router, the same swap, sometimes the same liquidity — just settled to Ethereum via a rollup batch instead of directly.

The math: an L2 transaction pays two fees in one — a tiny L2 execution component (cents) plus a fractional share of the L1 calldata-posting cost. When the rollup batches 10,000 transactions and posts them to Ethereum in one blob, your share of that L1 cost is one ten-thousandth of the full mainnet gas. EIP-4844 (blobs) dropped this further in March 2024 — most rollups now run 10-20x cheaper than they did pre-Dencun.

If you're doing anything routine — sending stablecoins, swapping, minting, bridging — start on an L2. Mainnet only when you need composability with mainnet-only contracts, are settling to a CEX that doesn't support L2 withdrawals, or are moving so much value that the bridge round-trip cost is dwarfed by the savings.

How to time the most expensive actions

Three actions in the table reliably cost more than “a normal transaction”: NFT mints (often 200k+ gas), Aave/Compound deposits (200k+), and contract deployments (600k+). For these, timing matters most. A mint at peak US trading hours can cost 5-10x the same mint at 4am UTC. Strategies:

• Check cheap gas times by day for the recurring weekly pattern.
• Use the gas fee forecast to spot the next lower-fee window in the upcoming 24 hours.
• For non-urgent expensive transactions, set a gas alert at your target gwei and wait.
• For mints and DeFi positions you'll close repeatedly, do them on an L2 — the 5-50x cost difference compounds.

Frequently asked questions