A mempool, short for memory pool, is the set of valid but unconfirmed transactions that a blockchain node holds in memory while they wait to be included in a block. When a user broadcasts a transaction, nodes check it, add it to their mempool and pass it on to their peers. Block producers then pick transactions from their own mempool, usually favouring those that pay the highest fees.
There is no single, official mempool. Every node keeps its own, shaped by what it has heard and by its local settings, so two mempools are similar but rarely identical. The mempool is where fee markets form, where wallets estimate how much to pay, and where many risks to unconfirmed payments arise. On Ethereum the same structure is often called the transaction pool or txpool.
A mempool in plain terms
Think of the waiting room at a busy clinic that sees patients in batches. Everyone who arrives gets checked at the door and takes a seat. Every few minutes a nurse calls in a batch, and patients who offered a higher fee for urgent care are called first. When the room is quiet, everyone gets in quickly; when it is packed, low payers can wait a long time, and if the room overflows, the lowest payers are asked to leave and come back later.
A mempool is that waiting room, except that there are thousands of branches of the clinic, each with its own waiting room and its own slightly different list of who is sitting there. The block producer is the nurse, and the fee offered decides the order.
How the mempool works
Before a node admits a transaction, it checks that the signatures are valid, the coins exist and are not already spent, and the transaction follows the node's relay policy (for example, a minimum fee rate and standard formats). Transactions that pass are added and relayed; the rest are dropped. A transaction that conflicts with one already in the mempool, trying to spend the same coins, is normally rejected, which offers limited protection against double spending but no guarantee.
When a new block arrives, the node removes every transaction it contains, plus any that now conflict with it. If a block becomes a stale block after a reorganisation, its transactions can be put back into the mempool.
Bitcoin Core, the most widely used Bitcoin software, limits the mempool to 300 MB by default and drops transactions that have waited longer than two weeks (336 hours). When the limit is reached, the transactions paying the lowest fee rate are evicted, and the minimum fee rate needed to enter rises until space frees up.
Fees and priority
Block space is scarce, so the mempool works as an auction. On the Bitcoin blockchain, miners rank transactions by fee rate, in satoshis per virtual byte, rather than by total fee, because a larger transaction uses more of the block size limit. The transaction fee page explains the general idea. Two tools help users who paid too little:
- Replace-by-fee (RBF). A sender broadcasts a new version of the transaction that pays a higher fee. BIP 125 introduced an opt-in version; since Bitcoin Core 28.0 in 2024, nodes accept replacements by default even without the opt-in signal.
- Child-pays-for-parent (CPFP). The recipient spends the stuck output in a new transaction with a high fee, so a miner who wants the child must include the parent too.
On the Ethereum blockchain each transaction sets a maximum fee and a priority fee (tip) per unit of gas; the protocol's base fee, which rises and falls with demand, is burned. A transaction cannot be mined until its maximum fee covers the base fee, and transactions from one account wait in order of their nonce. The pages on gas fee and gas price give details.
Congestion and what users see
When demand for block space exceeds supply, the mempool grows and fees climb. Bitcoin's mempool has swollen during market rallies and, in 2023, during the rush to inscribe data on the chain known as Ordinals, leaving low-fee payments unconfirmed for days. Mempool size and fee levels are shown live by sites such as mempool.space, a block explorer that specialises in this data, and wallets read similar data to suggest fees. Some congestion is relieved by moving activity to a layer 2 network such as the Lightning Network, which settles many payments in few on-chain transactions.
For merchants and exchanges, a transaction in the mempool is not yet a payment. Until it has confirmations, it can be replaced, evicted or outbid, so accepting a zero-confirmation transaction is a business risk that should be limited to small amounts.
Transparency, front-running and private mempools
Because the public mempool is visible to anyone who runs a node, pending transactions reveal what traders are about to do. On Ethereum, bots watch for large swaps on automated market makers and place their own trades just before and after them, a practice described as front-running or a "sandwich attack". The profit available from reordering, adding or excluding transactions is known as maximal extractable value (MEV).
In response, many Ethereum transactions now bypass the public mempool through private relays that pass them directly to block builders, and most validators buy ready-made blocks from specialised builders. Critics argue this concentrates power in a few builders and relays and makes the mempool less representative of real demand; supporters say it shields ordinary users from predatory bots. Similar debates exist on Bitcoin over which transactions node operators should relay and how miners should treat out-of-band fee payments.