diff --git a/mp-spdz-rs/src/fhe/plaintext.rs b/mp-spdz-rs/src/fhe/plaintext.rs
index dd65da6..14512b8 100644
--- a/mp-spdz-rs/src/fhe/plaintext.rs
+++ b/mp-spdz-rs/src/fhe/plaintext.rs
@@ -76,6 +76,18 @@ impl<C: CurveGroup> Plaintext<C> {
         self.inner.num_slots() as usize
     }
 
+    /// Create a plaintext given a batch of scalars
+    pub fn from_scalars(scalars: &[Scalar<C>], params: &BGVParams<C>) -> Self {
+        assert!(scalars.len() < params.plaintext_slots(), "not enough plaintext slots");
+
+        let mut pt = Self::new(params);
+        for (i, scalar) in scalars.iter().enumerate() {
+            pt.set_element(i, *scalar);
+        }
+
+        pt
+    }
+
     /// Get a vector of scalars from the plaintext slots
     pub fn to_scalars(&self) -> Vec<Scalar<C>> {
         let mut scalars = Vec::with_capacity(self.num_slots());
diff --git a/offline-phase/src/lowgear/input_masks.rs b/offline-phase/src/lowgear/input_masks.rs
new file mode 100644
index 0000000..985304e
--- /dev/null
+++ b/offline-phase/src/lowgear/input_masks.rs
@@ -0,0 +1,92 @@
+//! Generates input masks for a party
+
+use ark_ec::CurveGroup;
+use ark_mpc::{algebra::Scalar, network::MpcNetwork};
+use itertools::Itertools;
+use mp_spdz_rs::fhe::{ciphertext::Ciphertext, plaintext::Plaintext};
+use rand::rngs::OsRng;
+
+use crate::{error::LowGearError, structs::ValueMacBatch};
+
+use super::LowGear;
+
+impl<C: CurveGroup, N: MpcNetwork<C> + Unpin + Send> LowGear<C, N> {
+    /// Generate input masks for the given party
+    pub async fn generate_input_masks(&mut self, n: usize) -> Result<(), LowGearError> {
+        assert!(
+            n <= self.params.plaintext_slots(),
+            "can only generate input masks for {} slots",
+            self.params.plaintext_slots()
+        );
+
+        // Each party generates their values, shares, and mac shares
+        let mut rng = OsRng;
+        let my_values = (0..n).map(|_| Scalar::<C>::random(&mut rng)).collect_vec();
+        let my_share = (0..n).map(|_| Scalar::<C>::random(&mut rng)).collect_vec();
+
+        let mut mac_mask = Plaintext::new(&self.params);
+        mac_mask.randomize();
+        let my_key = self.mac_share;
+        let my_mac_shares =
+            my_values.iter().zip(mac_mask.to_scalars()).map(|(x, y)| my_key * x - y).collect_vec();
+
+        let my_values_shares = ValueMacBatch::from_parts(&my_share, &my_mac_shares);
+        self.input_masks.add_local_masks(my_values.clone(), my_values_shares.into_inner());
+
+        // Compute the counterparty's shares and mac shares of my values
+        let their_share = my_values.iter().zip(my_share.iter()).map(|(x, y)| x - y).collect_vec();
+        let other_key_enc = self.other_mac_enc.as_ref().unwrap();
+        let values_plaintext = Plaintext::from_scalars(&my_values, &self.params);
+        let mut mac_product = other_key_enc * &values_plaintext;
+        mac_product.rerandomize(self.other_pk.as_ref().unwrap());
+
+        let their_mac = &mac_product + &mac_mask;
+
+        // Exchange shares and macs
+        self.send_network_payload(their_share).await?;
+        let my_shares: Vec<Scalar<C>> = self.receive_network_payload().await?;
+
+        self.send_message(&their_mac).await?;
+        let my_counterparty_macs: Ciphertext<C> = self.receive_message().await?;
+        let mut my_macs = self.local_keypair.decrypt(&my_counterparty_macs).to_scalars();
+        my_macs.truncate(n);
+
+        let my_counterparty_shares = ValueMacBatch::from_parts(&my_shares, &my_macs);
+        self.input_masks.add_counterparty_masks(my_counterparty_shares);
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use ark_mpc::PARTY0;
+
+    use crate::test_helpers::mock_lowgear_with_keys;
+
+    /// Tests generating input masks
+    #[tokio::test]
+    async fn test_generate_input_masks() {
+        const N: usize = 100;
+        let (party0_res, _) = mock_lowgear_with_keys(|mut lowgear| async move {
+            lowgear.generate_input_masks(N).await.unwrap();
+
+            // Open the first party's input masks, verify that they're the same as party 0's
+            // cleartext values
+            if lowgear.party_id() == PARTY0 {
+                let (cleartext, shares) = lowgear.input_masks.get_local_mask_batch(N);
+                let opened = lowgear.open_and_check_macs(&shares).await.unwrap();
+
+                cleartext == opened
+            } else {
+                let shares = lowgear.input_masks.get_counterparty_mask_batch(N);
+                lowgear.open_and_check_macs(&shares).await.unwrap();
+
+                true
+            }
+        })
+        .await;
+
+        assert!(party0_res);
+    }
+}
diff --git a/offline-phase/src/lowgear/mod.rs b/offline-phase/src/lowgear/mod.rs
index ee6f9f4..5c25c78 100644
--- a/offline-phase/src/lowgear/mod.rs
+++ b/offline-phase/src/lowgear/mod.rs
@@ -2,6 +2,7 @@
 //! keys, authenticating inputs, etc
 
 pub mod commit_reveal;
+pub mod input_masks;
 pub mod inverse_tuples;
 pub mod mac_check;
 pub mod multiplication;
@@ -29,7 +30,7 @@ use rand::thread_rng;
 
 use crate::{
     error::LowGearError,
-    structs::{LowGearParams, LowGearPrep, ValueMacBatch},
+    structs::{InputMasks, LowGearParams, LowGearPrep, ValueMacBatch},
 };
 
 /// A type implementing Lowgear protocol logic
@@ -52,6 +53,12 @@ pub struct LowGear<C: CurveGroup, N: MpcNetwork<C>> {
     pub shared_bits: ValueMacBatch<C>,
     /// The shared random values generated during the offline phase
     pub shared_randomness: ValueMacBatch<C>,
+    /// The input masks generated during the offline phase
+    ///
+    /// An input mask is party specific, that is, each party has a set of input
+    /// values wherein it holds a random value in the cleartext
+    /// and the parties collectively hold a sharing of the value
+    pub input_masks: InputMasks<C>,
     /// A reference to the underlying network connection
     pub network: N,
 }
@@ -75,6 +82,7 @@ impl<C: CurveGroup, N: MpcNetwork<C> + Unpin> LowGear<C, N> {
             inverse_tuples: Default::default(),
             shared_bits: Default::default(),
             shared_randomness: Default::default(),
+            input_masks: Default::default(),
             network,
         }
     }
diff --git a/offline-phase/src/structs.rs b/offline-phase/src/structs.rs
index f08a93a..5c967be 100644
--- a/offline-phase/src/structs.rs
+++ b/offline-phase/src/structs.rs
@@ -4,7 +4,7 @@ use std::ops::{Add, Mul, Sub};
 
 use ark_ec::CurveGroup;
 use ark_mpc::algebra::{Scalar, ScalarShare};
-use ark_mpc::offline_prep::OfflinePhase;
+use ark_mpc::offline_prep::PreprocessingPhase;
 use mp_spdz_rs::fhe::ciphertext::Ciphertext;
 use mp_spdz_rs::fhe::keys::{BGVKeypair, BGVPublicKey};
 use mp_spdz_rs::fhe::params::BGVParams;
@@ -93,7 +93,7 @@ impl<C: CurveGroup> LowGearPrep<C> {
     }
 }
 
-impl<C: CurveGroup> OfflinePhase<C> for LowGearPrep<C> {
+impl<C: CurveGroup> PreprocessingPhase<C> for LowGearPrep<C> {
     fn next_shared_bit(&mut self) -> ScalarShare<C> {
         self.bits.split_off(1).into_inner()[0]
     }
@@ -172,6 +172,11 @@ impl<C: CurveGroup> ValueMacBatch<C> {
         self.inner.is_empty()
     }
 
+    /// Pop the last value and mac from the batch
+    pub fn pop(&mut self) -> Option<ScalarShare<C>> {
+        self.inner.pop()
+    }
+
     /// Append the given batch to this one
     pub fn append(&mut self, other: &mut Self) {
         self.inner.append(&mut other.inner);
@@ -296,6 +301,71 @@ impl<C: CurveGroup> Mul<&[Scalar<C>]> for &ValueMacBatch<C> {
     }
 }
 
+// ---------------
+// | Input Masks |
+// ---------------
+
+/// The input mask values held by the local party
+///
+/// Each party holds a set of random cleartext values used to mask inputs to the
+/// MPC. The other parties collectively hold a sharing of the values
+///
+/// So, this struct holds the local party's cleartext values and the local
+/// party's shares of their own and others' cleartext masks
+#[derive(Clone, Default)]
+pub struct InputMasks<C: CurveGroup> {
+    /// The local party's cleartext mask values
+    pub my_masks: Vec<Scalar<C>>,
+    /// The local party's shares of their own mask values
+    pub my_mask_shares: ValueMacBatch<C>,
+    /// The shares of the cleartext values
+    ///
+    /// Index `i` is a set of shares for party i's masks
+    pub their_masks: ValueMacBatch<C>,
+}
+
+impl<C: CurveGroup> InputMasks<C> {
+    /// Append values to `my_masks`
+    pub fn add_local_masks(&mut self, values: Vec<Scalar<C>>, masks: Vec<ScalarShare<C>>) {
+        assert_eq!(values.len(), masks.len());
+        self.my_masks.extend(values);
+        self.my_mask_shares.append(&mut ValueMacBatch::new(masks));
+    }
+
+    /// Add values to `their_masks`
+    pub fn add_counterparty_masks(&mut self, mut masks: ValueMacBatch<C>) {
+        self.their_masks.append(&mut masks);
+    }
+
+    /// Get the local party's next mask and share of the mask
+    pub fn get_local_mask(&mut self) -> (Scalar<C>, ScalarShare<C>) {
+        assert!(!self.my_masks.is_empty(), "no local masks left");
+        let mask = self.my_masks.pop().unwrap();
+        let mask_share = self.my_mask_shares.pop().unwrap();
+
+        (mask, mask_share)
+    }
+
+    /// Get a batch of local masks and shares of the masks
+    pub fn get_local_mask_batch(&mut self, num_masks: usize) -> (Vec<Scalar<C>>, ValueMacBatch<C>) {
+        let split_idx = self.my_masks.len() - num_masks;
+        let masks = self.my_masks.split_off(split_idx);
+        let mask_shares = self.my_mask_shares.split_off(num_masks);
+
+        (masks, mask_shares)
+    }
+
+    /// Get the local party's share of the counterparty's next mask
+    pub fn get_counterparty_mask(&mut self) -> ScalarShare<C> {
+        self.their_masks.split_off(1).into_inner()[0]
+    }
+
+    /// Get a batch of the local party's shares of the counterparty's masks
+    pub fn get_counterparty_mask_batch(&mut self, num_masks: usize) -> ValueMacBatch<C> {
+        self.their_masks.split_off(num_masks)
+    }
+}
+
 #[cfg(test)]
 mod test {
     use ark_mpc::{
@@ -305,7 +375,8 @@ mod test {
 
     use crate::test_helpers::mock_lowgear_with_triples;
 
-    /// Tests the use of the `LowGear` type as an `OfflinePhase` implementation
+    /// Tests the use of the `LowGear` type as an `PreprocessingPhase`
+    /// implementation
     #[tokio::test]
     async fn test_lowgear_offline_phase() {
         // Setup the mock offline phase